Cartographic Perspectives, Number 100, FORTHCOMING

4.0 International License. To view a copy of this license, visit http://creativecommons.org/licenses/by-nc-nd/4.0.

Visual Storytelling with Maps – Song et al. | 10

Zihan Song

Esri

zsong@esri.com

Robert E. Roth

University of Wisconsin–Madison

reroth@wisc.edu

Lily Houtman

The Pennsylvania State University

lhoutman@psu.edu

Timothy Prestby

The Pennsylvania State University

tprestby@psu.edu

Alicia Iverson

University of Wisconsin–Madison

aiverson3@wisc.edu

Song Gao

University of Wisconsin–Madison

gao@wisc.edu

Visual Storytelling with Maps: An Empirical Study on Story

Map Themes and Narrative Elements, Visual Storytelling

Genres and Tropes, and Individual Audience Differences

DOI: 10.14714/CP100.1759

PEER-REVIEWED ARTICLE

Visual storytelling describes the communication of stories through illustrations, graphics, imagery, and video instead of,

or in addition to, oral, written, and audio formats. Compared to their popularity and wide reach, empirical research

on map-based visual stories remains limited. We work towards inlling this gap through an empirical study on data

journalism, providing the rst assessment of four emerging design considerations for visual storytelling with maps: story

map themes and their constituent narrative elements, visual storytelling genres, visual storytelling tropes, and individ-

ual audience dierences. Specically, we recruited 125 participants to an online map study, requiring them to separately

review two visual stories and respond to a series of free-response and Likert scale questions regarding their retention,

comprehension, and reaction. We followed a 2×2×2 factorial design for the visual stories, varying their themes (US pres-

idential campaign donations, US coastal sea-level rise), genres (longform infographic, dynamic slideshow), and

tropes (color highlighting, leader lines), while holding other design dimensions constant. e story theme did not inu-

ence the participants’ total retention or comprehension, indicating that a three-act narrative and its constituent elements

can be applied consistently and eectively across variable kinds of topics. Instead, genres and, to a weaker degree, tropes

inuenced total participant retention, pointing to the importance of intentional design in map-based visual storytelling.

Overall, participants performed better when the visual storytelling designs used longform infographics or “scrollytell-

ing” (genres) to structure content and leader lines (tropes) to visually accent information. In contrast, the story theme

inuenced audience reaction, with participants feeling signicantly more concerned about and upset with the US presi-

dential campaign donations story compared to the US sea-level rise story. Individual audience dierences by expertise,

motivation, and prior beliefs also inuenced participant reaction. Our study signals a need for establishing a research and

education agenda on map-based visual storytelling in both cartography and data journalism.

KEYWORDS: visual storytelling; data journalism; spatial narratives; story maps; scrollytelling; narrative visualization

INTRODUCTION

H,    empirical research aimed at under-

standing how to design maps that support visual storytell-

ing. Mapping and storytelling have long been intertwined

(Denil 2017). e fourth-century Classic of Mountains and

Seas made visual the mythical story of ancient China; the

medieval Beatine Map was embedded within and rein-

forced the story of Christianity; Ogilby’s Britannia atlas of

1675 presented sequenced recollections of life on the road

in an increasingly interconnected Britain. While any sto-

ryteller can use maps, graphics, sketches, etc., to advance

Cartographic Perspectives, Number 100, FORTHCOMING Visual Storytelling with Maps – Song et al. | 11

their narrative and enhance their story, spatialized stories

within the margins of a single printed map or unfolded

through an atlas-like sequence of bound maps (Ormeling

1995). In the following, we use story to describe an account

of specic events, places, and people, and narrative to de-

scribe the structure and presentation of this content that

shapes the meaning of the story (Pearce 2008).

Arguably, both maps and stories are more accessible and

inuential today than ever due to pervasive computing,

innovations in new media, and advancing geoweb tech-

nologies (Haklay, Singleton, and Parker 2008; Sui and

Goodchild 2011; Sieber et al. 2016; Young, Hermida,

and Fulda 2018). Unsurprisingly, narrative and storytell-

ing have garnered substantial research attention at the

intersection of cartography, geography, and GIScience

(e.g., Elwood 2006; Pearce 2009; Phillips 2012; Caquard

2013) and increasingly are topics of inquiry in related, vi-

sual-centric elds such as information visualization and

visual analytics (e.g., Gershon and Page 2001; Eccles et

al. 2008; Ma et al. 2012; Kosara and Mackinlay 2013).

Professionally, the use of maps and graphics for storytell-

ing has become a dening trait of data journalism, or news

stories supplemented and even generated by analysis and

presentation of digital information (Gray, Chambers, and

Bounegru 2012). While journalists have a long history of

using data-driven maps as evidence in their news reports

(Monmonier 1989), many news organizations are explor-

ing novel narrative structures and design strategies as they

transition from a primarily print to a primarily digital me-

dium (Wallace 2016; Cairo 2017). In the following, we

adopt a broad denition of visual storytelling as the com-

munication of stories through illustrations, graphics, im-

agery, and video instead of or in addition to oral, written,

and audio formats (for a review of storytelling visualiza-

tions, see Segel and Heer 2010).

Despite both scholarly and practical advancement in the

history, application, and critique of narrative and story

in cartography and related elds, there remains relatively

limited empirical research on the intentional design of vi-

sual stories, particularly on map-based strategies and tech-

niques, and the subsequent interpretation of these designs

by their audiences. We addressed this gap through an em-

pirical study providing the rst assessment of four emerg-

ing design considerations for visual storytelling with maps:

story map themes and their constituent narrative elements,

visual storytelling genres, visual storytelling tropes, and

individual audience dierences. Specically, we asked:

1. What is the inuence of story map themes and their

constituent narrative elements on the audience’s

retention, comprehension, and reaction? Visual

stories covering dierent kinds of topics, or story

themes, still can share design similarities based

on the underlying narrative structure. A three-act

narrative—dating to Aristotle’s Poetics (ca. 335

BCE) and commonly adopted in play- and screen-

writing—comprises a set-up (Act 1), a conict/

confrontation (Act 2), and a resolution (Act 3) to

give the story a beginning, middle, and ending.

Each act includes recurring narrative elements

paced to build suspense through rising action and

then tie up loose threads through falling action.

e elements of a three-act narrative can inform

the selected sequence of maps and graphics for

a visual story, enforcing continuity to produce

a linear reading of inherently non-linear, often

two-dimensional, geographic information. For

our research, we designed two map-based visual

stories on timely topics seen in US media outlets,

using a consistent three-act narrative structure and

similar constituent narrative elements: the rst

about the inuence of US presidential campaign

donations on election results and a second about

the inuence of US coastal sea-level rise on climate

change vulnerability.

2. What is the inuence of visual storytelling genres

on the audience’s retention, comprehension, and re-

action? Broadly, a genre is a category of literature,

music, or other form of artistic expression that ex-

hibits similarity in structural and stylistic elements

(see Cartwright [1999] for the rst reference to

genres related to storytelling in cartography). Our

prior work has extended Segel and Heer (2010)

to identify seven visual storytelling genres made

possible by developments in pervasive comput-

ing, new media, and geoweb technologies (Roth

2021). ese genres dier by the visual or inter-

active techniques they use to enforce continuity of

elements in the narrative sequence. In the research

we present here, we examined dierences in the

audience’s retention, comprehension, and reaction

between two visual storytelling genres: longform

infographics and dynamic slideshows.

Cartographic Perspectives, Number 100, FORTHCOMING Visual Storytelling with Maps – Song et al. | 12

3. What is the inuence of visual storytelling tropes on

the audience’s retention, comprehension, and reac-

tion? A trope is a literary or rhetorical device used

to advance a story, much like a gure of speech

(Smith 1996). We previously established seven

visual storytelling tropes that capture a range of

design techniques used not to represent infor-

mation, but to advance the narrative and develop

narrative elements (Roth 2021). Employing visual

storytelling genres that enforce a linear, three-act

narrative utilizes a rst trope—continuity—by

unifying otherwise disparate visual elements into

a logical structure (Gershon and Page 2001). In

addition, we examined design techniques used to

focus attention—a second trope rst discussed

by Gershon and Page—on important or unusu-

al information that should not be missed by the

audience. Specically, we investigated dierences

in the audience’s retention, comprehension, and

reaction between two visual attention strategies

commonly used in cartography and information

visualization (e.g., Robinson 2011; Grin and

Robinson 2015): leader lines and color highlighting.

4. What is the inuence of individual audience

dierences on their retention, comprehension, and

reaction? Visual stories are presented from a situ-

ated perspective and invite the audience to draw

from their personal backgrounds and experiences

to derive meaning from the story (Pearce 2014).

Maps and stories are persuasive and political

(Harley 1989; Cronon 1992), and commonly are

employed together for controversial, divisive topics

(Vujaković 2014; Kent 2017). Multiple personal

characteristics can inuence retention, compre-

hension, and reaction, and therefore the success of

a visual story design. We term these variable audi-

ence characteristics individual dierences, and col-

lected information on expertise, motivation, and

prior beliefs on a number of topics related to visual

storytelling to examine the inuence of individual

audience dierences on retention, comprehension,

and reaction.

We addressed these research questions through an online

map study with 125 participants recruited from Amazon

Mechanical Turk. e study required participants to sep-

arately review two visual stories and then respond to a

series of multiple choice, free response, and Likert scale

questions to assess their retention, comprehension, and

reaction. We followed a 2×2×2 factorial design for our vi-

sual story materials, varying one of our test dimensions

(themes, genres, or tropes) while holding the others con-

stant, resulting in eight unique visual story designs in total.

e remainder of this paper describes related background

work, specics about our method design, overall results,

and concluding take-homes for the intentional design of

map-based visual stories.

RELATED WORK

“S ,” “ ,”  “

” are now commonplace terms in the car-

tographic lexicon, often evoked to simultaneously describe

a mode of individual expression, a visual design meth-

od, and a technological platform. Research on “narrative

cartography” is as diverse as that on maps themselves,

with storytelling opening new intellectual spaces for cin-

ematic (e.g., Caquard and Taylor 2009; Muehlenhaus

2014), imaginative (e.g., Joliveau 2009; Caquard 2011),

Indigenous (e.g., Chapin, Lamb, and Threlkeld 2005;

Pearce and Louis 2008), literary (e.g., Moretti 2005;

Bushell 2012), multimedia (e.g., Monmonier 1992;

Cartwright 1999), and participatory (e.g., Elwood 2006;

Miller 2006) mappings. Maps can give spatial structure

to oral, written, and audio-visual forms of storytelling

(Caquard and Cartwright 2014), and often are combined

with graphics, images, videos, and text to provide a deep

account of people, places, and events (Macfarlane 2007).

We explored map-based visual storytelling through the

lens of data journalism, an area that has seen increased re-

search and professional interest in cartography, informa-

tion visualization, and related elds (for recent edited vol-

umes, see Gray, Chambers, and Bounegru 2012; Riche et

al. 2018; Engebretsen and Kennedy 2020). Data journal-

ism is an iterative process that includes collecting disparate

data, analyzing and ltering the collected data, visualiz-

ing the data, and ultimately forming a story that hinges

upon key insights within the data (Weber and Rall 2012;

Rogers 2014). us, the data journalism process is much

like the highly iterative process we follow in cartography:

both journalist and cartographer are active in shaping an

Cartographic Perspectives, Number 100, FORTHCOMING Visual Storytelling with Maps – Song et al. | 13

explanation of the compiled text, graphics, and images. In

other words, both maps and news stories—and the various

combinations therein—exhibit purposeful design (Roth

2021). In our research, we approached four emerging

design considerations for visual storytelling using a case

study in data journalism.

STORY MAP THEMES AND THREE-ACT

NARRATIVE ELEMENTS

First, we drew from elements of a linear, three-act narra-

tive to inform the content and sequence of maps, graph-

ics, and text for two dierent visual storytelling themes

common in data journalism. Conceptually, nearly all data

journalism lends itself to mapping, as events occur in spe-

cic geographic, historical, and social contexts. Vujaković

(2014, 15) characterizes seven “news maps” themes and

18 sub-themes receptive to mapping in data journalism,

ranging from environmental concerns to politics. Each

story theme covers a unique knowledge domain and there-

fore may represent a dierent reporting responsibility in

a news room, with the themes broad enough to apply to

many geographic locations.

If the theme informs the content of the story, a linear nar-

rative provides design guidance for structuring and pre-

senting this story content. In this way, map-based visual

stories covering very dierent themes can share similar-

ities in their design if using the same narrative structure

(see Phillips 2012 for an analysis of common narrative

structures in geography), and consideration of the con-

stituent narrative elements of this structure during story

planning oers new opportunities for visual story design.

Specically, we organized narrative elements into a three-

act structure dening the beginning, middle, and end of

each of our story themes:

1. e set-up (Act 1) introduces the setting, key

characters, and problem context. e set-up often

includes a hook, or an exciting early scene that cap-

tures the attention of the audience and encourages

them to continue reading. For visual storytelling,

a map primarily frames the setting and problem

context in the set-up act, but places depicted with-

in the map also can be treated as exemplar protag-

onist or antagonist characters.

2. e conict or confrontation (Act 2) rst intervenes

with the problem, or key issue driving the story,

and then slowly builds suspense through rising

plot points. e problem produces tension among

characters, particularly between the protagonist

and antagonist for critical juxtaposition. e char-

acters respond and evolve at each plot point. In a

cartographic context, individual plot points can be

represented as either unique symbols and annota-

tions within a single map or unique maps within

a broader sequence of graphics, images, and text.

Accordingly, representation of a single narrative

element often is described as a frame within the

overall visual story (after Pearce 2008).

3. e resolution (Act 3) culminates the narrative arc

with the dramatic climax, or nal confrontation

between characters. e resolution concludes the

story with falling action in the denouement, in

which remaining matters are explained or re-

solved. Several narrative elements can be left unre-

solved for the audience in a clihanger, stimulating

their imagination and curiosity while allowing

them to “ll in the gaps” using their own experi-

ences and predictions.

There are a number of modifications and extensions of

a three-act narrative (see Hullman et al. 2013; öny et

al. 2018), and visual storytelling often deviates from a

linear narrative to temporarily withhold information or

build suspense (Muehlenhaus 2014). However, we test-

ed two dierent visual story themes instead of two dif-

ferent narrative structures, both to simplify the factorial

study design described below and to mitigate the eect of

participant biases towards any single theme, which oth-

erwise might skew our results when testing other visual

story design considerations (i.e., genres and tropes). We

selected the case studies of US presidential campaign dona-

tions and US coastal sea-level rise to exemplify Vujaković’s

(2014) “Politics, Internal” versus “Environment and

Science” themes, two timely topics in US media outlets.

Table 1 denes the aforementioned constituent elements

of a linear, three-act narrative based on Roth (2021) and

describes their application to the two case studies used in

the online study.

VISUAL STORYTELLING GENRES

Second, we examined the influence of the visual story-

telling genre on participants’ retention, comprehension,

and reaction. Segel and Heer (2010, 1139) proposed seven

basic “genres of narrative visualization,” differentiating

Cartographic Perspectives, Number 100, FORTHCOMING Visual Storytelling with Maps – Song et al. | 14

Narrative

Element

Description Story 1: US Presidential Campaign Donations Story 2: US Sea-level Rise Vulnerability

Theme

The general thematic

category of the visual story

(Vujakovic 2014)

A. Politics: internal; 1. Government

D. Environment and Science; 8. Environmental problems/

impacts

Topic

The specic geographic

phenomenon or process

covered in the visual story

US presidential campaign donations US sea-level rise

Title

(Panel 1)

A condensed, engaging

headline for the visual story

The Presidency’s Price Tag: Campaign Donations and the

2012 Presidential Election

Soaking in Water: Sea-Level Rise and Vulnerable Coastal

Properties Since 2012

Summary

A brief introduction to the

visual story following a three-

act narrative structure

Purpose: This story follows two swing states—Colorado

and Ohio—to explain the impact of campaign donations

on the US presidential election results. Problem:

Differences in party campaign donations inuenced

voting results in many swing states during the 2012

Presidential Election. Resolution: Colorado and Ohio

represent different alternatives for addressing campaign

donations.

Purpose: This story follows two coastal states—New

York and North Carolina—to explain the impact of

sea-level rise on the vulnerability of coastal properties

in the US. Problem: Rising sea levels have increased the

vulnerability of properties on the East Coast of US since

2012. Resolution: New York and North Carolina represent

different alternatives for addressing sea-level rise.

Act 1: Set-up

Setting

The specic place, time, and

social context, giving the

story a geography

Space Where the story takes place US Swing States US Eastern Coastal States

Time When the story occurred The 2012 US Presidential Election The 2012 Hurricane Season

Characters

The people or places who

embody the narrative and

act-out the plot

Protagonist

(accented)

The main character in the

story

Colorado: A swing state whose voting support increased

for the Democratic candidate as Democrats gained an

advantage in campaign donations

New York: A coastal state whose vulnerability increased

as sea level rose

Antagonist

(accented)

The character in opposition

to the protagonist

Ohio: A swing state whose voting support was largely not

inuenced by an advantage in campaign donations by

either party

North Carolina: A coastal state whose vulnerability was

largely not inuenced by rising sea levels

The Hook

(Panel 1)

An exciting early scene that

captures the audience’s

interest and encourages them

to continue reading

Private donations, not public discourse, shape the

outcome of the presidential election

Even small rises in sea-level dramatically increase coastal

vulnerability to storms

Problem

Context

(Panel 2)

Additional background

information needed to

interpret the story later in the

narrative sequence

Title: What Is Happening with Our Elections? It Starts

with Rising Campaign Costs. Fact: US presidential

campaign costs have increased nearly 800% in the past

40 years. Accent: Campaign costs peaked at $1.74

billion in the 2008 presidential election. Graph: Y value:

Presidential Election Costs ($ Billion); X value: Year.

Title: What is Happening with Our Coasts? It Starts with

Rising Temperatures. Fact: US average temperatures

have increased almost 3 °F in the past 40 years. Accent:

US average temperatures peaked at 54.3 °F in 2015.

Graph: Y value: Average Temperature (°F); X value:

Year.

Problem

Context

(Panel 3)

Additional background

information needed to

interpret the story later in the

narrative sequence

Title: Why Do Costs Matter? More than 50% of

Campaign Funds were from Donations in 2012.

Colorado: The average person in Colorado donated

$3.30 during the 2012 presidential election. Ohio: The

average person in Ohio donated only $1.50 during the

2012 presidential election. Legend: title: Presidential

Campaign Donations; description: Average donations per

person ($), 2012 presidential election.

Title: Why Do Coasts Matter? More than 50% of US

Citizens Lived in Coastal Areas by 2012. New York:

3,081 people live in an average square mile of New

York coasts. North Carolina: Only 73 people live in an

average square mile of North Carolina coasts. Legend:

title: Coastal Population Density; description: Average

people per square mile of coastal area, 2012.

Table 1. Elements of a Three-act Narrative. A linear, three-act narrative comprises a set-up (Act 1), conict/confrontation (Act 2), and

resolution (Act 3). This table describes how we applied the constituent elements of a linear, three-act narrative to the pair of visual stories

used in this study. Continued on the next page.

Cartographic Perspectives, Number 100, FORTHCOMING Visual Storytelling with Maps – Song et al. | 15

each genre by the number and order of frames within the

story: magazine style, annotated chart, partitioned post-

er, ow chart, comic strip, slide show, and lm/video/an-

imation. While foundational, the original Segel and Heer

taxonomy drew primarily upon printed news maps and

passive television news reporting, thus preceding many

emerging design practices made possible by pervasive

computing, new media, and geoweb technologies (Kosara

and Mackinlay 2013). Further, the number of frames

is less relevant with these emerging technologies, where

page space is unlimited. In response, we previously pro-

posed a revised taxonomy of visual storytelling genres

based only on the visual or interactive technique used to

enforce continuity of elements in the narrative sequence:

Act 2: Conict/Confrontation

Problem/

Catalyst

(Panel 4)

The central confrontation,

obstacle, or setback driving

the story

Title: So What? Increasing Donations Pose A Problem…

Fact: The Democrat advantage in campaign donations

reached $253 million for the 2012 presidential election.

Accent: Democrats received $51 million more donations

than Republicans in September, the largest donations

advantage during the 2012 president election. Graph:

Y value: Cumulative Donation Gap ($ Million); X value:

Month

Title: So What? Rising Sea Levels Pose a Problem…

Fact: The US average sea levels in 2012 reached 47.8

millimeters above the 2002 average. Accent: Global

sea levels rose 8.4 millimeters in 2012, the largest sea-

level rise from 2002-2012. Graph: Y value: Cumulative

Sea Level Change (Millimeters); X value: Year.

Tension

(Panel 5)

The impact of the problem

on the protagonist versus the

antagonist

Title: …Particularly for Swing States. Colorado:

Democrats increased their support by 6.5% in the swing

state of Colorado. Ohio: Democrats only increased their

support by 0.7% in the swing state of Ohio. Legend:

title: Increase in Voting Lead; description: Change in

Democratic lead (% total), July 2012 poll to Nov 2012

election.

Title: …Particularly for States on the East Coast. New

York: The average value of vulnerable properties

in coastal areas is $24,800 in New York. North

Carolina: The average value of vulnerable properties in

coastal areas is only $7,730 in North Carolina. Legend:

title: Property Vulnerability; description: Average value of

vulnerable coastal properties ($), 2012

Plot Points

(Panel 6)

One in a sequence of events

motivated by the problem

that impacts the characters

(Cause)

Title: A Deeper Look: Democrats Gained their Largest

Donation Advantage in Major Cities. Fact: Democrats

drew 140% more urban-based donations per person in

Colorado than Ohio, largely attributed to the progressive

Denver metro area. Accent: The Democrats gained an

advantage of $4,400,000 in Denver, the highest urban

lead in swing states; Legend: title: Donation Gap

description: Democrat advantage in campaign donations

($), 2012

Title: A Deeper Look: Sea Levels Rose the Most in Major

Stations. Fact: Sea-level rose 50% more in urban-centers

in New York compared to North Carolina, particularly

due to the dense infrastructure in New York City and

Long Island. Accent: Sea-level annual rate in Bergen

Point is 4.4 millimeters, the highest rate among stations

in coastal states. Legend: title: Average Sea-Level Rise

description: Average annual sea-level rise (millimeters/

year), 1992–2012.

Act 3: Resolution

Climax

(Panel 7)

The nal plot point bringing

characters together to face

their tension and consider

competing solutions (Effect)

Title: As a Result, Campaign Donations Have a Different

Inuence on Election Results in Swing States like Colorado

versus Ohio. Colorado: Every $100 advantage for the

Democrats bought 7.5 votes in Colorado. Ohio: Every

$100 advantage for the Democrats bought only 2.8 votes

in Ohio.

Title: As a Result, Sea-level Rise Has a Different Impact

on Vulnerability in Coastal States like New York versus

North Carolina. New York: Every inch in sea-level

rise exposes $3,900 of property in New York. North

Carolina: Every inch in sea-level rise only exposes

$1,400 of property in North Carolina.

Resolution

& Denoue-

ment

(Panel 8)

Falling action in which all

remaining matters with

the setting, characters,

and problem context are

explained or resolved

Title: What’s Next? Colorado and Ohio Represent

Different Alternatives for Addressing Campaign

Donations. Colorado: Colorado has imposed new

regulations to limit campaign funding since the 2012

presidential election. Ohio: At the same time, Ohio has

failed to act on campaign funding.

Title: What’s Next? New York and North Carolina

Represent Different Alternatives for Addressing Sea-level

Rise. New York: New York has invested considerable

public funds to prevent sea-level rise related crises.

North Carolina: At the same time, North Carolina has

failed to act on sea-level rise.

Cliffhanger

(Panel 9)

The dramatic ending,

leaving open strands for the

audience to ponder

Title: What Do You Think We Should Do As a Nation?

Colorado: The Democrats are predicted to make

only a 2.3% gain in Colorado in the 2020 presidential

election if campaign funding remains consistent from

2016. Ohio: The Republicans are predicted to make a

whopping 12.5% gain in Ohio in the 2020 presidential

election if campaign funding remains consistent from the

2016. Legend: title: Voting Results Predictions, 2020

Presidential Election; description: Predicted Republican

voting lead; Predicted Democratic voting lead

Title: What Do You Think We Should Do as a Nation?

New York: The value of vulnerable properties per

person in New York is predicted to increase to $290 by

2020 if sea levels continue to increase at a consistent

rate. North Carolina: The value of vulnerable

properties per person in North Carolina is predicted to

increase a surprising $325 by 2020 if sea levels continue

to increase at a consistent rate. Legend: title: Predicted

Property Vulnerability; description: Increased value

of vulnerable coastal properties per person ($), 2020

Prediction

Table 1 (continued). Elements of a Three-act Narrative. A linear, three-act narrative comprises a set-up (Act 1), conict/confrontation (Act

2), and resolution (Act 3). This table describes how we applied the constituent elements of a linear, three-act narrative to the pair of visual

stories used in this study.

Cartographic Perspectives, Number 100, FORTHCOMING Visual Storytelling with Maps – Song et al. | 16

static visual stories (encapsulating most of the Segel and

Heer taxonomy), longform infographics, dynamic slide-

shows (from Segel and Heer), narrated animations (from

Segel and Heer), multimedia visual experiences, person-

alized story maps, and compilations (Roth 2021; Table 2).

We also imagine the ability to mash up these genres, com-

bining dierent visual or interactive techniques for enforc-

ing continuity within a single visual story.

Of these possible genres, we narrowed our focus for this

research to longform infographics and dynamic slide-

shows. Longform infographics enforce continuity through

vertical reading and browser scrolling, with the genre often

described as “scrollytelling” by data journalists (Stolper

et al. 2016, 8; citing Bostock 2014). In contrast, dynam-

ic slideshows enforce continuity by advancement through

a series of visual panels or frames of consistent size and

format, producing a discrete, typically horizontal scroll

versus the continuous vertical scroll of longform info-

graphics; this discrete sequencing of content sometimes is

described as “pagination” when used as a web design strat-

egy (Wieczorek et al. 2014, 310). We selected these two

genres for our initial investigation in order to contrast the

slideshow presentations common across academia, gov-

ernment, and industry (Kosara and Mackinlay 2013) with

the longform infographic or “scrollytelling” approaches

now common in news media, as this structural dierence

in the method for enforcing continuity between the two

genres (and across all genres) potentially influences the

audiences’ retention, comprehension, and reaction. For

instance, eye-tracking studies have shown that scrollable

web content promotes visual skimming (Nielsen 2006),

a potential disadvantage of longform infographics com-

pared to dynamic slideshows. In contrast, dynamic slide-

shows explicitly dose information into less complex slides

to reduce skimming, an advantage previously found to aid

retention in text presentations (Wieczorek et al. 2014).

However, the scrolling in longform infographics en-

ables continuous, audience-driven pacing (Harrower and

Sheesley 2005), and thus makes the user more active in

the experience. In contrast, dynamic slideshows require

discrete advancement of slides at a potentially monotonous

pace that is designer-driven, which may be exacerbated by

lags from reloading wrapper page content for each slide.

We provide additional discussion on the relative advantag-

es and limitations of dierent visual storytelling genres in

Roth (2021).

VISUAL STORYTELLING TROPES

ird, we considered new visual design techniques—de-

scribed as tropes—used not to represent information, but

rather to enhance the narrative. Our use of tropes syn-

thesized disparate literature on narrative cartography and

visualization. For instance, Gershon and Page (2001, 34)

described “story-like visual presentation,” listing story-

telling concepts like conflict and ambiguity resolution,

continuity, eective redundancy, lling gaps, increasing

Genre Linearity Denition

Static Visual Story Linearity is enforced through partitioning of the layout into frames and clarifying annotation.

Longform Infographic Linearity is enforced through vertical reading and browser scrolling.

Dynamic Slideshow Linearity is enforced by advancement through a series of slides.

Narrated Animation Linearity is enforced by the progression of digital display time.

Multimedia Visual Experience Linearity is enforced by anchor tags and hyperlinking.

Personalized Story Map Linearity is enforced by the order that an individual contributes content to the map.

Visual Story Compilation Linearity is enforced by unfolding events in near real-time or major updates to the design.

Table 2. Visual Storytelling Genres. Visual storytelling genres are dened by the visual or interactive technique used to enforce linearity of

elements in the narrative sequence. This table describes the taxonomy of visual storytelling genres introduced by Roth (2021), expanded

from Segel and Heer (2010). We examined the longform infographic and dynamic slideshow genres in this study.

Cartographic Perspectives, Number 100, FORTHCOMING Visual Storytelling with Maps – Song et al. | 17

attention, and setting the mood.

Similarly, Pearce (2009, 4) com-

bined “narrative techniques” like

ambiguity, closure, focalization,

scale, and voice to express “qual-

ities of place: intimacy, identity,

and connection with the read-

er.” Elsewhere, such tropes are

described collectively as story-

telling aordances (Kosara and

Mackinlay 2013), devices (Segel

and Heer 2010), or figures

(Cattoor and Perkins 2014). As

described in our prior work, we

organized these existing con-

cepts and recommendations into

seven design tropes for visual

storytelling: continuity (central

to the delineation of the dif-

ferent genres described above),

mood, dosing, attention, re-

dundancy, metaphor, and voice

(Roth 2021; Table 3). Each

trope then has an associated set

of design techniques that can be

employed for visual storytelling.

In addition to dierences in continuity by genre, we ex-

amined dierent methods for focusing attention on im-

portant or unusual information in the story that should

not be missed (Gershon and Page 2001). Attention as a

trope describes a range of design solutions that produce

an “Isolation Eect,” making one item stand out over oth-

ers in a visual scene (Lidwell, Holden, and Butler 2010,

254). Commonly used techniques for focusing attention

in cartography and visualization include highlighting fea-

tures through the visual variables (e.g., Robinson 2011);

applying annotations such as leader lines, flow arrows,

appended geometric frames, opacity masks, numbering,

call-outs, and labels (e.g., Pearce 2008); and creating dy-

namic changes through blinking/flickering, panning/

zooming, and focus+context visualizations (e.g., Weber

Reuschel, Piatti, and Hurni 2014). We provided addi-

tional discussion of these focusing attention techniques in

Roth (2021) using the catch-all term visual accenting. In

the presented study, we specically examined the dier-

ences in attention between leader lines and color highlight-

ing within a single representation, similar to Grin and

Robinson’s (2015) investigation into the use of these two

visual accenting techniques for focusing attention between

two coordinated representations. Notably, Griffin and

Robinson found leader lines to perform equally well as the

more common color highlighting for focusing attention

between coordinated views. We held all other visual story-

telling tropes constant across tested materials.

INDIVIDUAL DIFFERENCES

Finally, we addressed the role of individual differences

among the audience on the eectiveness of visual story-

telling. Just like maps, stories and their meanings are not

xed or objective truths, but rather are shaped by the situ-

ated backgrounds and experiences of both storyteller and

audience (after Haraway 1991; Rose 1997). While our

contribution in this paper is to the design of map-based

visual stories, and new ways that a cartographer or data

journalist can shape the narrative, we recognize that the

same design will conjure different understandings and

evoke dierent responses from dierent people in dier-

ent places (Pearce 2014). An advantage to taking a nar-

rative approach to cartography is that it enables the em-

brace of pluralism—a data feminism principle—allowing

designers to be more transparent in their positionality

Table 3. Visual Storytelling Tropes. Visual storytelling tropes are design techniques used

not to represent information, but rather to enhance the narrative. This table summarizes

emerging tropes identied by Roth (2021). Two visual accenting solutions for focusing

attention were examined in this study: color highlighting and leader lines, following Grifn

and Robinson (2015).

Trope Denition

Continuity Unify otherwise disparate visual elements into a logical structure.

Mood Set a visual tone congruent with the narrative and its elements.

Dosing

Reduce overall complexity of story content into incremental chunks of

information.

Attention

Emphasize important or unusual information that cannot be missed in

the story.

Redundancy Repeat important or unusual information to develop story themes.

Metaphor

Bring together seemingly unrelated concepts in a single frame to

facilitate understanding of complex narrative elements.

Voice

Embed situated experiences, opinions, and values into the visual story

to clarify meaning.

Cartographic Perspectives, Number 100, FORTHCOMING Visual Storytelling with Maps – Song et al. | 18

toward the story theme, while also inviting the audience

to form multiple, equally-valid story meanings (adapted

from D’Ignazio and Klein 2020). Considering the individ-

ual audience dierences that inuence visual storytelling

can help cartographers and data journalists make more in-

formed choices about their designs (Fish 2020b).

To this end, we drew upon work on mapping context and

individual dierences in cartography and visualization (see

Grin et al. 2017 for a comprehensive review and syn-

thesis). Expanding upon our prior work (Roth 2009), we

rst collected several measures of our participants’ exper-

tise—including their education, training, familiarity, and

interest—regarding topics related to map-based visual

storytelling. ese included their experience with dier-

ent news media (print and online), maps and information

graphics, computing technology and the internet, and rel-

evant news story themes.

We then collected measures of motivation regarding the

same topics, an individual difference we have found to

be as important as expertise for the success of web maps

(Roth and Harrower 2008).

Finally, we assessed prior beliefs developed from past ex-

perience with topics related to the visual stories that may

bias a reader in favor for or against a contentious posi-

tion. Regarding map-based visual stories, prior beliefs

may persist even in the face of evidence that invalidates

them (Cohen 2012). Both tested themes followed Phillip’s

(2012) cause and eect three-act narrative structure, and

we embedded a subtle conservative lean in the US presi-

dential campaign donations visual story and a subtle liberal

lean in the US coastal sea-level rise visual story to balance

prior beliefs for the other visual story design consider-

ations. We did not assess the inuence of other sociode-

mographic dierences, given the sensitivity of the visual

story themes and the focused goals of the research.

METHODS

PARTICIPANTS

O  - participants completed

an online map study assessing their retention, compre-

hension, and reaction across dierent map-based visual

story designs. We recruited participants from Amazon

Mechanical Turk in March 2017—after rst conducting a

pilot survey with four participants in a controlled environ-

ment in the University of Wisconsin Cartography Lab to

capture potential issues with the survey design. e pilot

study resulted in small text and styling changes, but no

major changes to the study design.

Recruitment using Mechanical Turk has both advantag-

es and limitations for our study (see Hauser, Paolacci,

and Chandler 2019 for expanded discussion). We chose

Mechanical Turk over alternatives, such as university stu-

dent recruitment pools, to capture greater demographic,

geographic, and political diversity among participants,

which was important for our ability to assess the inuence

of individual differences and, specifically, prior beliefs,

on participants’ retention, comprehension, and reaction.

However, recruitment with Mechanical Turk restricted

us from purposefully sampling by individual dierences

without the risk of dramatically oversampling one group;

accordingly the sample variability in individual dierenc-

es may have an inuence on the study results. Additional

limitations include variable participant attention and expe-

rience with research studies, which we mitigated partially

through questions on individual dierences and balanced

experimental procedure. Finally, Mechanical Turk has

prompted new ethical considerations for human subjects

research regarding the exploitation of labor and associated

expected quality of experimental results (D’Ignazio and

Klein 2020). We designed the survey to take 30 minutes

to complete, and participants received $4 USD for com-

pensation, a rate that exceeded the Wisconsin minimum

wage at the time of the study ($7.25/hr USD). Participants

completed the survey with a median time of 30.71 and av-

erage of 34.64 minutes.

Of the 125 participants in our sample, 71 identified as

male and 54 as female, with zero responses to non-binary

categories, and an average age of 35 years old. Nineteen

participants did not attend college, 43 attended some col-

lege or were attending college, 55 completed an under-

graduate degree, six completed a graduate degree, and two

reported “Other.”

Online participants completed the survey on their own

computing devices and were instructed to use non-mobile

devices. Sixty-three participants completed the survey on

a laptop computer and 62 on a desktop computer. Because

Cartographic Perspectives, Number 100, FORTHCOMING Visual Storytelling with Maps – Song et al. | 19

Figure 1. Participant characteristics.

Cartographic Perspectives, Number 100, FORTHCOMING Visual Storytelling with Maps – Song et al. | 20

all evaluated map-based visu-

al stories dealt with US-based

issues (see below), we limited

participation to the US gener-

al public. Figure 1 provides an

overview of participant char-

acteristics; the Supplemental

Materials include an additional

breakdown of the sample by in-

dividual differences by experi-

mental condition.

MATERIALS

e online map study followed a

2×2×2 factorial design (Montello

and Sutton 2012):

• Factor 1: emes.

Conditions consisted of

stories on US presidential

campaign donations and US

sea-level rise, exemplifying

Vujaković’s (2014) “Politics,

Internal” and “Environment and Science” themes.

Both conditions followed Phillip’s (2012) cause and

eect three-act narrative structure and used real

data for their contents, with US presidential campaign

donations receiving a subtle conservative lean in the

cause and eect structure and US sea-level rise a subtle

liberal lean to balance eects from prior beliefs.

• Factor 2: Genres. Conditions consisted of longform

infographics and dynamic slideshows. We rst dosed

story content for both conditions into nine-frame

storyboards (three frames per act)—with each frame

representing a dierent narrative element—and then

assembled the frames using the continuity technique

dening the given genre (continuous scrolling versus

discrete slide advancement, respectively) to promote

information equivalency between conditions. We

loaded all nine frames onto the survey webpage for

the longform infographics condition, although frames

were spaced so that only one was visible at a time

while scrolling up to a 1080p resolution screen in

order to mimic “lazy loading” of content that was

conventional in web design at the time of the study.

In contrast, only one frame was loaded into the

webpage at a time for the dynamic slideshow condition,

again following convention at the time of the study of

AJAX (Asynchronous JavaScript and XML) or sepa-

rate webpages to load content on demand. e result

was an intentional delineation in the experience be-

tween continuous and discrete scrolling. Participants

could return to prior frames in both genres (i.e.,

scroll-up, previous slide), but were prevented from

returning to the visual stories after advancing past the

last frame.

• Factor 3: Tropes. Conditions consisted of two common

design solutions for focusing attention, leader lines and

color highlighting. We used black for color highlighting,

as black was not used for symbolization and therefore

could be applied consistently across the graphics. e

use of black also avoided any potential issues with

color vision deciency in the participant sample.

For design consistency, leader lines also used black

strokes. e attention solutions were applied to the

protagonist and antagonist in each design, leading to

the introduction of two complementary place-based

characters for each story.

The 2×2×2 factorial design resulted in construction of

eight unique stories (Table 4). e overall aesthetic style

Story Theme (Factor 1) Genre (Factor 2) Trope (Factor 3)

Story 1 US sea-level rise Longform Infographic Color highlighting

Story 2

US presidential

campaign donations

Longform Infographic Leader lines

Story 3

US presidential

campaign donations

Longform Infographic Color highlighting

Story 4 US sea-level rise Longform Infographic Leader lines

Story 5

US presidential

campaign donations

Dynamic Slideshow Leader lines

Story 6 US sea-level rise Dynamic Slideshow Color highlighting

Story 7 US sea-level rise Dynamic Slideshow Leader lines

Story 8

US presidential

campaign donations

Dynamic Slideshow Color highlighting

Table 4. Factorial Design. The study followed a 2×2×2 factorial design, resulting in eight

unique visual stories.

Cartographic Perspectives, Number 100, FORTHCOMING Visual Storytelling with Maps – Song et al. | 21

and layout (i.e., mood), the

amount and complexity of

story content (i.e., dosing),

and the visualization de-

signs remained consistent

for comparability. Figure

2 provides an overview of

the two, nine-frame visual

stories and Figure 3 pro-

vides a preview of the sur-

vey interface for advancing

both genres. All tested ma-

terials are included in the

Supplemental Materials.

PROCEDURE

e online map study began

with an overview of project

purpose and goals. After

obtaining consent, partici-

pants completed a training

block to combat learning

effects. The training block

contained the opening three

frames of a third example

visual story on US book-

store sales that followed the

same design rules of other

materials and included sim-

ilar questions as the other

experimental blocks (also

included in Supplemental

Materials). Participants

were allotted as much time

as needed to review the

training block before pro-

gressing to the experimental

blocks.

Participants then complet-

ed two experimental blocks

covering the US presidential

campaign donations and US

sea-level rise themes (Table

5). We assigned Factors

1 (themes) and 3 (tropes)

within-subjects, with as-

signment balanced so that

Figure 2. Visual Story Design. Both visual story themes were designed as a nine-panel sequence—

with each panel representing a different narrative element (Table 1)—that could be presented as

either a longform infographic or dynamic slideshow. Top: The US presidential campaign donations

theme using leader lines. Bottom: The US sea level rise theme using color highlighting. High

resolution versions of all tested materials are available as supplemental materials.

Cartographic Perspectives, Number 100, FORTHCOMING Visual Storytelling with Maps – Song et al. | 22

each possible combination of themes and tropes was

viewed by approximately one quarter of the study partici-

pants. We assigned Factor 2 (genres) between-subjects so

that participants viewed just one linear structure during

the survey, limiting confusion over layout and interface

changes to the survey instrument between the two blocks.

Participants responded to questions on separate survey

pages after reviewing the visual story in full. Common

measures of map design effectiveness include accuracy/

correctness, response time, and completeness (Sweeney,

Maguire, and Shackel 1993). However, Kosara and

Mackinlay (2013) and Figueiras (2014) argue that the

eectiveness of visual stories should instead be based on

engagement and interest when reading the story, the abil-

ity to remember key points in the story, and the ability to

make better informed decisions after reading the story.

Finally, we have advocated in prior work to evaluate visu-

al stories as much on their “eectiveness” as on how they

make the audience feel about the depicted geographies

(Roth 2021). Accordingly, participants responded to three

kinds of questions about the story within each block, with

each set of questions its own survey page:

1. Retention: e retention page included 12 multiple

choice questions built from benchmark mapping

tasks (Roth 2013). We included these questions to

supply baseline “accuracy” or “correctness” metrics

common in empirical research on map design out-

side of visual storytelling (after Sweeney, Maguire,

and Shackel 1993). e questions varied on the

level of predicted diculty: three ordinal com-

pare tasks (easiest), three ordinal rank tasks, and

six numerical identify tasks (hardest). All tasks

were phrased at an elementary map reading level

to focus attention onto specic narrative elements

in the visual story (Andrienko, Andrienko, and

Gatalsky 2003). e twelve questions were ran-

domized to avoid learning eects between blocks.

We placed the retention questions at the end of

the experimental block to reduce short-term recall,

with retention questions following the compre-

hension and reaction questions.

2. Comprehension: Participants summarized the

content of the story through a single, open-ended

question, which we added to capture qualitative

Figure 3. Survey Design. The survey design used two different

interactive techniques to advance frames based on the examined

genre. Top: The longform infographic condition loaded all frames

into the webpage and used browser scrolling by mouse wheel

or side scroll bar to advance frames, spaced to show only one

frame at a time. Bottom: The dynamic slideshow condition loaded

one frame at a time and used bottom “Back” and “Next” buttons

to advance frames.

Table 5. Group Assignment. Each participant reviewed both

visual story themes, with the visual stories presented in the same

genre but using different trope solutions.

Group

Sample Size

(n=125 total)

First

Viewed

Second

Viewed

Group I 33 Story 1 Story 2

Group II 30 Story 3 Story 4

Group III 31 Story 5 Story 6

Group IV 31 Story 7 Story 8

Cartographic Perspectives, Number 100, FORTHCOMING Visual Storytelling with Maps – Song et al. | 23

and potentially meaningful engagement and

interest with the visual story (after Kosara and

Mackinlay 2013; Figueiras 2014). We instructed

participants to be as comprehensive as possible and

to use their own words when phrasing their sum-

mary. We then coded the open-ended respons-

es following tenets of qualitative data analysis

(Caudle 2004), with codes based on the narrative

elements shaping the visual stories: setting (space,

time), characters (protagonist, antagonist), prob-

lem, cause, eect/resolution, and clihanger, as

well as a nal code to capture any mistakes in

comprehension (e.g., see Table 1 for example state-

ments that would be coded as “comprehended”).

us, the coding scheme measured whether pur-

poseful treatment of these narrative elements in

the visual story design resulted in reader compre-

hension. e codes were binary based on presence

or absence within the participant open-ended

response (i.e., did the participant comprehend the

narrative element correctly in their response). e

rst 10% of the comprehension responses were

coded by two, independent coders to hone code

denitions and clarify ambiguity in the coding

scheme.

3. Reaction: Participants responded to a series of

seven-point Likert scales to capture how they felt

about the visual stories and their depicted plac-

es and people (after Roth 2021). Reaction scales

included participant interests in and beliefs about

the visual story. Participants also self-reported

their core aect in reaction to the visual story,

including audience arousal (activated vs. deactivat-

ed) and hedonic valence (pleasant vs. unpleasant;

Grin and McQuoid 2012).

Participants completed this set of questions for both ex-

perimental blocks, resulting in responses to 250 unique

blocks through the sample of 125 participants.

Participants concluded the study with an exit survey to

characterize individual dierences that might inuence

retention, comprehension, and reaction. e exit survey

included questions on expertise, motivation, and prior be-

liefs using ordinal Likert scales. e complete online map

study protocol is available in the Supplemental Materials.

ANALYSIS

Each factor (themes, genres, tropes) in the study design

served as an independent variable, with responses to reten-

tion, comprehension, and reaction serving as dependent

variables, respectively, and individual dierences as inter-

action eects. We applied factorial ANOVA to assess the

inuence of the three factors on retention, comprehension,

and reaction, as well as the pairwise interaction eects be-

tween factors to establish independence. We used factori-

al ANOVA instead of individual t-tests to mitigate alpha

accumulation (Type I error) across a large number of hy-

pothesis tests. We prepared a separate factorial ANOVA

model for each unique retention (Table 6), comprehen-

sion (Table 7), and reaction (Table 8) measure, resulting

in 23 total factorial ANOVA models. Figures 4, 5, and 6

visually illustrate dierences by conditions for the reten-

tion, comprehension, and reaction measures, respectively,

marking dierences by themes, genres, or tropes identied

as signicant through factorial ANOVA.

We then used multiple linear regression (MLR) models to

assess the inuence of individual dierences on retention

and comprehension (Table 9). We used MLR over a series

of Spearman rank correlationsagain to mitigate alpha ac-

cumulation. MLR quanties the relationship of predictors

to a single response variable in the form of B weight coe-

cients (Nathans, Oswald, and Nimon 2012). We chose the

B weight, which is unstandardized, to assess variable im-

portance because each predictor variable was measured in

the same units (Allen 2017). A coecient of zero indicates

the predictor has no influence on the response variable

while a positive/negative coecient indicates that, with all

other variables held constant, for every increase of one unit

in the predictor, the response variable increases/decreas-

es on average the value by of the coecient respectively.

Since some questions capturing individual differences

were specic to one story over the other, we ran three dif-

ferent MLR models for comprehension and retention: re-

sponses to the US presidential campaign donations condition

alone, the US sea-level rise condition alone, and pooled re-

sponses together for individual dierences relevant to both

conditions.

We used a version of ordinal linear regression (OLR)

known as the proportional odds model (POM) to analyze

the reaction Likert scales (Brant 1990; Table 10). POM is

a non-parametric method that quanties the proportional

odds ratio (θ) between predictor variables and the response

Cartographic Perspectives, Number 100, FORTHCOMING Visual Storytelling with Maps – Song et al. | 24

variable (McCullagh 1980). e range of the odds ratio is

from 0 to positive innity, with every one-point increase

in the Likert scale predictor having x times (odds ratio)

the odds of the response variable being a unit higher. For

instance, a one unit increase in a predictor with an odds

ratio of 1.4 increases the odds of the response variable

being one unit higher by 40%. We ran eight models test-

ing the inuence of individual dierences on each of the

eight reaction Likert scales.

We also tested for multicollinearity among measures of

individual differences to reduce the variance/standard

error in the regression coecient estimates. We calculat-

ed the variance inuence factors (VIFs) for all individual

dierences and removed variables with VIFs over 4 to re-

duce multicollinearity, as VIF=4 means that the standard

error for the coecient of that predictor variable is 2 times

(i.e., the square root of 4) larger than if that predictor vari-

able had 0 correlation with the other predictor variables

(Lavery et al. 2019). We removed seven of the collected

individual dierence measures as a result, which are high-

lighted in yellow in the results tables.

In the following discussion, we present retention results

rst, as comprehension and reaction enrich the quantita-

tive analysis on retention. We present interaction eects

from individual dierences last.

Table 6. Participant Retention Results. The table shows descriptive statistics (top) and factorial ANOVA (bottom) for retention. The table

includes main effects by factor (theme, genre, and trope) as well as interaction effects between factors. The table includes four separate

factorial ANOVA models on retention for compare (ordinal), rank (ordinal), identify (numerical), and total retention. Color indicates

signicance: p < 0.10 , p < 0.05 , p < 0.01 , p < 0.001.

Factor /

Interactions

 Compare (Ordinal) Rank (Ordinal) Identify (Numerical) Total Retention

Descriptive Statistics n  Mean SD  Mean SD  Mean SD  Mean sd

Total 3000 81.9% 24.5% 71.7% 2 7. 4 % 66.2% 26.8% 71.4% 20.7%

US presidential

campaign donations

1500 84.3% 27. 0 % 63.5% 27. 6% 65.3% 27.2% 69.6% 21.6%

US sea-level rise 1500 79.5% 21.5% 80.0% 24.7% 67.1% 26.5% 73.3% 19.6%

Longform

infographics

1512 84.9% 21.3% 70.9% 2 7. 6 % 71.0% 24.4% 74. 5% 18.2%

Dynamic slideshows 1488 78.8% 2 7. 0 % 72.6% 27.2% 61.3% 28.3% 68.3% 22.6%

Leader lines 1500 84.0% 24.5% 73.3% 2 7. 4 % 69.2% 26.8% 73.8% 20.7%

Color highlighting 1500 79.7% 25.4% 70.1% 2 7. 4% 63.2% 27.1% 69.1% 20.9%

Factorial ANOVA df

Mean

F p

Mean

F p

Mean

F p

Mean

F p

Theme 1 1.30 2.44 0.12 15.38 25.12 0.00 0.68 0.27 0.60 12.10 2.03 0.16

Genre 1 2.13 4.01 0.05 0.16 0.26 0.61 21.35 8.52 0.00 33.75 5.67 0.02

Trope 1 0.97 1.83 0.18 0.73 1.19 0.28 8.22 3.28 0.07 20.93 3.51 0.06

Theme : Genre 1 0.05 0.09 0.76 0.77 1.25 0.26 1.30 0.52 0.47 0.13 0.02 0.88

Theme : Trope 1 0.21 0.40 0.53 2.46 4.01 0.05 3. 74 1.49 0.22 14.72 2.47 0.12

Genre : Trope 1 0.18 0.33 0.57 0.02 0.03 0.85 0.63 0.25 0.62 0.13 0.02 0.88

Residuals 243 0.53   0.61   2.51   5.96  

Cartographic Perspectives, Number 100, FORTHCOMING Visual Storytelling with Maps – Song et al. | 25

Table 7. Participant Comprehension Results (in two parts). The table shows descriptive statistics (top half of each part) and factorial

ANOVA (bottom half) for comprehension. The table includes main effects by factor (theme, genre, and trope) as well as interaction effects

between factors. The table includes nine separate factorial ANOVA models on comprehension for each of the nine evaluated narrative

elements and two additional models for total comprehension across all elements and mistakes in comprehension. Color indicates

signicance: p < 0.10 , p < 0.05 , p < 0.01, p < 0.001.

Factor /

Interactions

 Space Time Protagonist Antagonist Problem

Descriptive n  mean sd  mean sd  mean sd  mean sd  mean sd

Total 250 26.5% 44.2% 18.0% 38.5% 59.6% 49.2% 52.4% 50.0% 96.0% 19.6%

US presidential

campaign donations

125 35.2% 48.0% 24.0% 42.9% 52.0% 50.2% 48.8% 50.2% 96.0% 19. 7%

US sea-level rise 125 17. 7 % 38.4% 12.0% 32.6% 6 7. 2 % 4 7.1% 56.0% 49.8% 96.0% 19.7 %

Longform

infographics

126 32.5% 4 7. 0 % 18.3% 38.8% 60.3% 49.1% 50.8% 50.2% 9 9.2% 8.9%

Dynamic slideshows 124 20.3% 40.4% 17. 7 % 38.4% 58.9% 49.4% 54.0% 50.0% 92.7% 26.0%

Leader lines 125 24.0% 42.9% 19.2% 39.5% 60.8% 49.0% 52.0% 50.2% 95.2% 21.5%

Color highlighting 125 29.0% 45.6% 16.8% 37. 5 % 58.4% 49.5% 52.8% 50.1% 96.8% 17. 7 % 

Factorial ANOVA df

Mean

F p

Mean

F p

Mean

F p

Mean

F p

Mean

F p

Theme 1 1.94 10.49 0.00 0.90 6.23 0.01 1.44 6.02 0.01 0.32 1.27 0.26 0.00 0.00 1.00

Genre 1 0.96 5.19 0.02 0.00 0.01 0.92 0.01 0.06 0.82 0.07 0.26 0.61 0.26 6.82 0.01

Trope 1 0.17 0.92 0.34 0.03 0.19 0.66 0.05 0.20 0.66 0.00 0.01 0.92 0.02 0.42 0.52

Theme : Genre 1 0.58 3.13 0.08 0.47 3.24 0.07 0.20 0.83 0.36 0.00 0.02 0.89 0.02 0.40 0.53

Theme : Trope 1 0.06 0.34 0.56 0.26 1.82 0.18 0.05 0.19 0.66 0.00 0.00 1.00 0.00 0.00 0.99

Genre : Trope 1 0.00 0.00 0.98 0.13 0.88 0.35 0.18 0. 76 0.38 0.01 0.02 0.88 0.00 0.00 0.98

Residuals 243 0.18   0.14   0.24   0.26   0.04  

Factor /

Interactions

 Tension Cause Effect Cliffhanger Total Mistakes

Descriptive n  mean sd  mean sd  mean sd  mean sd  mean sd  mean sd

Total 250 69.2% 46.3% 32.4% 46.9% 50.8% 50.1% 26.8% 44.4% 48.0% 22.9% 12.4% 33.0%

US Presidential

campaign donations

125 71.2% 45.5% 20.8% 40.8% 44.0% 49.8% 25.6% 43.8% 46.4% 22.7% 16.0% 36.8%

US sea-level rise 125 6 7. 2 % 4 7.1% 44.0% 49.8% 5 7. 6 % 49.6% 28.0% 45.1% 49.6% 23.0% 8.8% 28.4%

Longform

infographics

126 69.8% 46.1% 35.7% 48.1% 50.8% 50.2% 2 7. 0 % 44.6% 49.4% 21.7% 11.1% 31.6%

Dynamic slideshows 124 68.5% 46.6% 29.0% 45.6% 50.8% 50.2% 26.6% 44.4% 46.6% 24.0% 13.7% 34.5%

Leader lines 125 73.6% 44.3% 35.2% 48.0% 50.4% 50.2% 25.6% 43.8% 48.4% 23.2% 13.6% 34.4%

Color highlighting 125 64.8% 48.0% 29.6% 45.8% 51.2% 50.2% 28.0% 45.1% 4 7. 6 % 22.6% 11. 2% 31.6%

Factorial ANOVA df

Mean

F p

Mean

F p

Mean

F p

Mean

F p

Mean

F p

Mean

F p

Theme 1 0.10 0.46 0.50 3.36 16.36 0.00 1.16 4.63 0.03 0.04 0.18 0.67 5.18 1.21 0.27 0.32 2.98 0.09

Genre 1 0.01 0.05 0.83 0.28 1.36 0.25 0.00 0.00 1.00 0.00 0.00 0.95 3.93 0.92 0.34 0.04 0.39 0.53

Trope 1 0.47 2.19 0.14 0.24 1.15 0.28 0.00 0.01 0.94 0.03 0.17 0.68 0.47 0.11 0.74 0.03 0.29 0.59

Theme:Genre 1 0.11 0.52 0.47 0.32 1.56 0.21 0.11 0.43 0.51 0.10 0.48 0.49 0.25 0.06 0.81 0.03 0.30 0.58

Theme:Trope 1 0 .11 0.50 0.48 0.34 1.63 0.20 0.57 2.29 0.13 0.12 0.58 0.45 0.09 0.02 0.89 0.08 0.69 0.41

Genre:Trope 1 0.00 0.02 0.90 0.25 1.20 0.27 0.03 0.12 0.73 0.31 1.55 0.21 0.68 0.16 0.69 0.21 1.96 0.16

Residuals 243 0.22   0.21   0.25   0.20   4.30   0.11  

Cartographic Perspectives, Number 100, FORTHCOMING Visual Storytelling with Maps – Song et al. | 26

Table 8. Participant Reaction Results (in two parts). The table shows descriptive statistics (top half of each part) and factorial ANOVA

(bottom half) for reaction. Likert ratings are out of 7 points. The table includes main effects by factor (theme, genre, and trope) as well as

interaction effects between factors. The table includes eight separate factorial ANOVA models on reaction, one for each of the evaluated

reaction measures. Color indicates signicance: p < 0.10 , p < 0.05 , p < 0.01, p < 0.001.

Factor / Interactions  Interest+ (Interest) Interest- (Concern) Belief+ (Agree) Belief- (No Inuence)

Descriptive n  mean sd  mean sd  mean sd  mean sd

Total 2000 5.0 1.7 4.5 1.9 5.0 1.6 3.8 1.9

US presidential

campaign donations

1000 5.1 1.6 4.9 1.8 5.2 1.5 3.6 1.9

US sea-level rise 1000 4.8 1.8 4.0 1.9 4.8 1.6 4.0 1.8

Longform infographics 1008 5.1 1.6 4.4 1.9 5.2 1.5 3.6 1.8

Dynamic slideshows 992 4.9 1.8 4.5 1.9 4.8 1.6 4.0 2.0

Leader lines 1000 5.2 1.7 4.6 1.8 5.1 1.6 3.6 1.9

Color highlighting 1000 4.8 1.7 4.3 1.9 4.9 1.5 3.9 1.9

Factorial ANOVA df

Mean

F p

Mean

F p

Mean

F p

Mean

F p

Theme 1 4.62 1.62 0.21 46.66 14.01 0.00 10.82 4.57 0.03 7. 4 0 2.10 0.15

Genre 1 1.61 0.56 0.45 0.48 0.14 0.71 7. 74 3.27 0.07 11. 05 3.14 0.08

Trope 1 12.19 4.26 0.04 5.60 1.68 0.20 3.79 1.60 0.21 5.77 1.64 0.20

Theme : Genre 1 0.08 0.03 0.87 2.35 0.71 0.40 0.00 0.00 0.97 7. 6 6 2.17 0.14

Theme : Trope 1 1.87 0.65 0.42 8.28 2.49 0.12 4.34 1.83 0.18 0.62 0.18 0.68

Genre : Trope 1 13.79 4.81 0.03 3.74 1.12 0.29 0.08 0.03 0.85 2.06 0.58 0.45

Residuals 243 2.86   3.33   2.37   3.52  

Factor / Interactions  Arousal+ (Excite) Arousal- (Bore) Hedonic+ (Enjoy) Hedonic- (Upset)

Descriptive n  mean sd  mean sd mean sd  mean sd

Total 2000 3.5 1.7 3.1 1.9 4.4 1.8 3.2 1.8

U.S. presidential

campaign donations

1000 3.5 1.7 3.0 1.9 4.4 1.7 3.5 2.0

U.S. sea-level rise 1000 3.5 1.7 3.1 1.9 4.3 1.8 2.9 1.7

Longform infographics 1008 3.3 1.6 2.9 1.8 4.4 1.7 2.9 1.8

Dynamic slideshows 992 3.7 1.7 3.3 2.0 4.3 1.8 3.5 1.9

Leader lines 1000 3.6 1.7 2.9 1.8 4.5 1.7 3.2 1.9

Color highlighting 1000 3.4 1.6 3.2 1.9 4.2 1.8 3.1 1.8

Factorial ANOVA df

Mean

F p

Mean

F p

Mean

F p

Mean

F p

Theme 1 0.04 0.01 0.91 0.48 0.14 0.71 0.58 0.19 0.67 23.72 7. 2 4 0.01

Genre 1 7. 7 3 2.76 0.10 9.67 2.78 0.10 0.34 0.11 0.74 21.52 6.57 0.01

Trope 1 3.83 1.37 0.24 6.01 1.72 0.19 4.55 1.47 0.23 0.92 0.28 0.60

Theme : Genre 1 0.16 0.06 0.81 2.30 0.66 0.42 0.19 0.06 0.80 1.60 0.49 0.49

Theme : Trope 1 0.50 0.18 0.67 4.72 1.35 0.25 2.21 0.72 0.40 0.99 0.30 0.58

Genre : Trope 1 4.78 1.71 0.19 5.16 1.48 0.23 9.08 2.94 0.09 0.63 0.19 0.66

Residuals 243 2.80   3.49   3.09   3.28  

Cartographic Perspectives, Number 100, FORTHCOMING Visual Storytelling with Maps – Song et al. | 27

Table 9. Inuence of Individual Differences on Retention and Comprehension. The table shows the results of multiple linear regression

between individual differences and retention and comprehension. Color indicates signicance: p < 0.10 , p < 0.05 , p < 0.01, p < 0.001.

Yellow indicates an individual difference measure that we removed because it was collinear with another measure.

Individual Difference A: Inuence on Retention B: Inuence on Comprehension

Sea-level rise

Presidential

campaign

donations

Both Sea-level rise

Presidential

campaign

donations

Both

Relation to Visual Story Theme

Expertise:

Familiarity

Environment and Science -0.28 0.24 0.08 0.68

Domestic Politics 0.38 0.13 -0.09 0.69

Expertise:

Training

Environment and Science - 0.15 0.39 -0.42 0.01

Domestic Politics -0.06 0.77 - 0.12 0.52

Motivation:

Prior Interest

Environment and Science 0.04 0.84 -0.18 0.29

Domestic Politics -0.28 0.24 0.20 0.33

Prior Beliefs

Socially liberal-v-conservative 0.26 0.08 - 0.18 0.15

Fiscally liberal-v-conservative 0.20 0.19 0.00 0.98

Environmentally conscious-v-agnostic -0.37 0.03 -0.34 0.02

Politically active-v-agnostic -0.22 0.24 -0.05 0.76

Concern about sea-level rise -0.47 0.01 -0.44 0.01

Concern about coastal vulnerable properties

Concern about presidential campaign donations 0.07 0.69

Concern about the presidential election results - 0.12 0.40

I believe sea-level rise is a topic worth discussing 0.39 0.08

I believe presidential campaign donation is a topic worth

discussing

0.10 0.62 0.41 0.03

I believe sea-level rise is a problem 0.29 0.10

I believe presidential election campaign donation is a

problem

Relation to Design and Technology

Expertise:

Familiarity

Print News Sources 0.23 0.11 0.31 0 .11 0.57 0.01 - 0.16 0.29 - 0.13 0.45 -0.32 0.10

Online News Sources -0.65 0.01 -0.77 0.01 -1.20 0.00 -0.58 0.01 -0.56 0.02 -0.89 0.00

Maps -0.35 0.27 -0.26 0.27 -0.51 0.04 0.02 0.92 - 0.12 0.55 0.12 0.59

Computing Technology -0.00 0.59 - 0.13 0.59 -0.37 0.17 -0.22 0.25 -0.21 0.31 -0.57 0.02

The Internet 0.56 0.00 0.96 0.00 1.51 0.00 0.18 0.47 0.38 0.17 0.71 0.03

Information Graphics 0.34 0.72 0.08 0.72 0.45 0.09 0.24 0.18 0.28 0.17 0.44 0.07

Expertise:

Training

Print News Sources - 0.13 0.43 0.23 0.26 -0.05 0.82 -0.02 0.89 0.22 0.20 0.04

Online News Sources 0.826

Maps - 0.13 0.47 -0.47 0.02 -0.63 0.00 0.12 0.44 -0.09 0.63 -0.25

Computing Technology 0.20

The Internet 0.26 0.07 - 0.13 0.42 0.06 0.71 0.25 0.04 0.11 0.41 0.16

Information Graphics 0.304

Motivation:

Prior Interest

Print News Sources -0.32 0.09 -0.53 0.01 -0.80 0.00 -0.27 0.09 - 0.31 0.09 -0.37 0.07

Online News Sources 0.55 0.01 0.58 0.01 1.07 0.00 0.63 0.00 0.39 0.05 1.05 0.00

Maps 0.08 0.67 0.22 0.26 0.28 0.20 0.04 0.81 0.03 0.85 -0.04 0.86

Computing Technology -0.07 0.74 -0.01 0.97 0.16 0.55 - 0.11 0.54 -0.04 0.85 0.06 0.81

The Internet 0.14 0.59 0.26 0.37 0.37 0.26 0.12 0.57 0.13 0.60 0.31 0.29

Information Graphics 0.22 0.29 0.23 0.31 0.32 0.20 -0.27 0.12 - 0.21 0.27 -0.48 0.03

Cartographic Perspectives, Number 100, FORTHCOMING Visual Storytelling with Maps – Song et al. | 28

Table 10. Inuence of Individual Differences on Reaction. The table shows the results of multiple linear regression between individual

differences and reaction. Color indicates signicance: p < 0.10 , p < 0.05 , p < 0.01, p < 0.001. Yellow indicates an individual

difference measure that we removed because it was collinear with another measure.

Individual Difference

Arousal+

The visual story excited me.

Arousal-

The visual story bored me.

Hedonic+

I enjoyed the visual story.

Hedonic-

I was upset by the visual story.

Interest+

The visual story interested me.

Interest-

The visual story concern me.

Belief+

I agree with the visual story.

Belief-

The visual story did not

inuence my views.

Relation to Visual Story Theme

Expertise:

Familiarity

Environment and Science OR Domestic

Politics

0.93 0.60 1.33 0.03 0.80 0.09 0.76 0.03 0.71 0.01 0.73 0.02 0.81 0.11 1.39 0.01

Expertise:

Training

Environment and Science OR Domestic

Politics

1.15 0.19 1.25 0.03 1.07 0.49 1.08 0.50 0.98 0.85 1.10 0.33 1.09 0.40 1.12 0.28

Motivation:

Prior Interest

Environment and Science OR Domestic

Politics

1.11 0.38 0.89 0.29 1.02 0.85 1.37 0.01 1.15 0.24 1.09 0.46 0.92 0.46 0.94 0.59

Prior Beliefs

Socially liberal-v-conservative OR Fiscally

liberal-v-conservative

0.99 0.92 1.06 0.47 1.15 0.08 1.18 0.03 1.04 0.59 1.10 0.26 0.93 0.35 0.97 0.67

Environmentally conscious-v-agnostic OR

Politically active-v-agnostic

1.10 0.33 1.14 0.18 1.25 0.02 0.98 0.80 1.09 0.37 1.07 0.53 1.14 0.11 1.06 0.56

Concern about sea-level rise OR Concern

about presidential campaign donations

1.07 0.48 1.13 0.20 1.10 0.32 1.18 0.09 1.05 0.60 1.33 0.00 1.24 0.03 0.91 0.33

Concern about coastal vulnerable

properties

      

Concern about the presidential election

results

      

I believe sea-level rise is a topic worth

discussing OR I believe presidential

campaign donation is a topic worth

discussing

1.16 0.20 0.77 0.02 1.26 0.04 1.14 0.22 1.41 0.00 1.23 0.07 1.20 0 .11 0.78 0.03

I believe sea-level rise is a problem       

I believe Presidential election campaign

donation is a problem

      

Relation to Design and Technology

Expertise:

Familiarity

Print News Sources 0.98 0.88 0.95 0.58 0.99 0.90 1.02 0.90 1.10 0.34 0.94 0.55 0.93 0.50 1.06 0.61

Online News Sources 1.02 0.91 0.92 0.57 1.50 0.01 1.00 0.98 1.15 0.36 0.81 0.15 1.16 0.30 1.19 0.24

Maps 0.56 0.00 1.17 0.23 0.73 0.02 1.24 0.09 0.81 0.11 1.11 0.41 0.92 0.53 1.12 0.40

Computing Technology 0.83 0.16 1.22 0 .11 0.94 0.64 1.25 0.07 1.00 0.99 1.03 0.85 0.79 0.07 1.21 0.12

The Internet 0.91 0.57 0.95 0.77 0.77 0.13 0.88 0.42 0.80 0.19 1.05 0.76 0.96 0.82 0.84 0.29

Information Graphics 1.20 0.15 0.94 0.56 0.90 0.44 0.89 0.34 1.01 0.95 1.15 0.25 1.33 0.02 0.69 0.00

Expertise:

Training

Print News Sources 0.86 0.15 1.01 0.90 0.87 0.16 1.13 0.24 0.94 0.53 1.03 0.77 0.93 0.48 1.01 0.89

Online News Sources       

Maps 1.26 0.04 0.80 0.05 1.04 0.70 0.92 0.45 1.06 0.59 1.01 0.90 1.06 0.58 0.89 0.26

Computing Technology       

The Internet 0.98 0.81 1.18 0.06 1.00 1.00 1.01 0.87 1.00 0.97 0.96 0.61 1.00 0.98 1.11 0.21

Information Graphics  

    

Motivation:

Prior Interest

Print News Sources 1.38 0.00 0.99 0.94 1.15 0.21 1.12 0.34 1.13 0.26 1.18 0.15 1.18 0.14 0.87 0.20

Online News Sources 1.06 0.64 0.95 0.70 1.11 0.41 0.90 0.40 1.03 0.81 1.03 0.79 0.93 0.55 0.93 0.54

Maps 1.20 0.09 0.75 0.01 1.41 0.00 0.91 0.39 1.42 0.00 1.10 0.36 1.27 0.02 0.93 0.47

Computing Technology 1.02 0.88 0.78 0.04 1.00 1.00 0.84 0.16 1.14 0.30 1.16 0.23 1.18 0.20 0.81 0.09

The Internet 0.89 0.48 0.96 0. 74 1.04 0.80 0.87 0.36 1.15 0.39 0.95 0.77 1.56 0.01 0.85 0.32

Information Graphics 1.35 0.01 0.96 0.70 1.40 0.01 1.06 0.65 0.96 0.76 0.96 0.76 0.80 0.06 1.04 0.77

Cartographic Perspectives, Number 100, FORTHCOMING Visual Storytelling with Maps – Song et al. | 29

Figure 4. Retention Scores by Factorial Conditions. Color indicates signicance: p < 0.10, p < 0.05 , p < 0.01, p < 0.001. Size is used

redundantly with color shading to indicate signicance.

Figure 5. Comprehension Scores by Factorial Conditions. Color indicates signicance: p < 0.10, p < 0.05 , p < 0.01, p < 0.001. Size is

used redundantly with color shading to indicate signicance. Figure continues on the next page.

Cartographic Perspectives, Number 100, FORTHCOMING Visual Storytelling with Maps – Song et al. | 30

Figure 5 (continued). Comprehension Scores by Factorial Conditions. Color indicates signicance: p < 0.10, p < 0.05 , p < 0.01, p < 0.001.

Size is used redundantly with color shading to indicate signicance.

Cartographic Perspectives, Number 100, FORTHCOMING Visual Storytelling with Maps – Song et al. | 31

Figure 6. Reaction Scores by Factorial Conditions. Color indicates signicance: p < 0.10, p < 0.05 , p < 0.01, p < 0.001. Size is used

redundantly with color shading to indicate signicance.

Cartographic Perspectives, Number 100, FORTHCOMING Visual Storytelling with Maps – Song et al. | 32

RESULTS AND DISCUSSION

OVERALL RESULTS

P   

with an overall accuracy of 71.4%, suggesting that the

retention questions were neither obvious nor impossible

(Table 6). As expected, participants performed the best on

the simpler compare (ordinal) tasks (81.9%) and worst on

the more dicult to remember identify (numerical) tasks

(66.2%), with the accuracy for rank (ordinal) tasks in the

middle (71.7%).

Regarding comprehension, participants on average de-

scribed 48.0% of the narrative elements included in the

comprehension coding scheme (Table 7), a relatively rich

discussion about the tested stories given the general au-

dience recruited from Mechanical Turk and open-ended

format of the comprehension question. In contrast, only

12.4% of the responses included a mistake in comprehen-

sion. us, comprehension results generally showed that

three-act narrative structure was eective for map-based

visual storytelling. ere were several notable patterns in

overall comprehension that provide insight into the visual

design of elements in a three-act narrative. Nearly all par-

ticipants (96.0%) clearly stated the problem of the story in

their comprehension response, suggesting that the prob-

lem was the most salient narrative element in both visual

story themes. Although an expected nding, conrmation

that readers focused on the problem is useful for cartogra-

phers and data journalists, as this element should be em-

phasized in the story title and redundantly accented as a

central motif throughout the story. e three next most

commonly mentioned codes related to the characters in

the story: 69.2% of participants described the tension—or

the how the problem impacts the key characters—59.6%

described the protagonist, and 52.4% described the antag-

onist. is suggests that the focus attention strategies ap-

plied to the characters were relatively successful (although

dierentially so by condition, as described in the results by

tropes below). is description also provided evidence that

readers can conceptualize places or regions as characters in

an explicitly geographic story, presenting an opportunity

to add dramatic narrative structuring to map designs in vi-

sual stories. While both stories used Phillip’s (2012) cause

and eect narrative structure, the cause (32.4%) was dis-

cussed much less frequently than the eect (50.8%), show-

ing a bias towards outcomes compared to drivers in visual

story comprehension. Finally, spatial setting (26.5%), the

clihanger (26.8%), and temporal setting (18.0%) were

described infrequently. Given that the story was map-

based, it was surprising that the geographic location and,

to a lesser degree, the timeframe were not discussed more

in the comprehension responses, although it is possible

that the visual accenting of specic place-based characters

diverted attention away from the overall spatiotemporal

context.

e two strongest reaction registers to the visual stories

were on interest and agreement, receiving an average re-

sponse of 5.0/7 for both measures (Table 8). us, overall

the visual stories captured participant interest—demon-

strating the value of employing maps and graphics for

visual storytelling—and participants generally agreed

with the cause/eect narrative used to structure the story.

Participants also provided slightly positive ratings for their

concern about the story topic (4.5/7), but nearly neutral

ratings about the story’s inuence on them (3.8/7). ere

was marginal overall affective impact by arousal, with

positive arousal (excitement; 3.5/7) and negative arous-

al (boredom; 3.1/7) both receiving less than the “neither

agree nor disagree” midpoint of 4.0 out of 7. ere was

a slight positive aective response on the hedonic scale,

with participants finding the experience more pleasant

(enjoyment; 4.4/7) than unpleasant (upsetting; 3.2/7).

STORY MAP THEMES (FACTOR 1)

Factor 1 included two conditions by visual storytelling

theme: US presidential campaign donations and US sea-level

rise. Starting with retention (Table 6), there was no statis-

tical dierence between themes regarding total retention

(F = 2.03, p = 0.16). us, the theme did not impact the

“accuracy” or “correctness” of reading the maps and graph-

ics in the visual story, common metrics used in empirical

map design research. is is a potentially exciting nd-

ing for cartographers and data journalists, suggesting that

the visualized theme may not inuence story retention for

a general audience if the design follows a three-act nar-

rative structure. Looking at specic retention questions,

there were no signicant dierences between themes with

regard to the easy compare (ordinal) tasks (F = 2.44, p =

0.12) and the dicult identify (numerical) tasks (F = 0.27,

p = 0.60). However, there was a significance difference

Cartographic Perspectives, Number 100, FORTHCOMING Visual Storytelling with Maps – Song et al. | 33

between themes in retention for the rank (ordinal) tasks

(F = 25.12, p = 0.00) as well as a signicant interaction ef-

fect between theme and tropes for rank (ordinal) tasks (F =

4.01, p = 0.05). e interaction possibly could be attributed

to the phrasing of sea-level “rise” for the US sea-level rise

condition, which may have suggested an increasing trend,

thus requiring fewer visual cues to focus attention on to

the highest ranked year (see Supplemental Materials).

Similarly, there was no statistical dierence in total com-

prehension between themes in the average number of

narrative elements discussed (F = 1.21, p = 0.27; Table 7),

adding evidence that the visualized theme may not in-

uence performance for a general audience if the design

follows a three-act narrative structure. However, partic-

ipants made more mistakes in comprehension for the US

presidential campaign donations condition (16.0%) than the

US sea-level rise condition (8.8%), a signicant dierence

at alpha = 0.10 (F = 2.98, p = 0.09), potentially pointing to

prior misconceptions about the 2016 US presidential elec-

tion. ere also were several notable dierences in com-

prehension between themes for specic narrative elements.

Participants discussed the spatial (F = 10.49, p = 0.00) and

temporal (F = 6.23, p = 0.01) settings signicantly more

frequently in their open-ended responses for the US pres-

idential campaign donations story than the US sea-level rise

story. In contrast, participants discussed the cause (F =

16.36, p = 0.00) and eect (F = 4.63, p = 0.03) more fre-

quently for the US sea-level rise story than the US presiden-

tial campaign donations story. While there may be a variety

of explanations for these dierences by specic narrative

elements, we suspect that the setting (“swing states”) was

implicitly more central to understanding the US presi-

dential campaign donations story and the cause-eect re-

lationship (“climate change”) was implicitly more central

to understanding the US sea-level rise story, despite con-

trolling the information and design complexity for these

narrative elements in both visual stories. us, while the

theme did not inuence total comprehension, individual

themes and unique visual story designs of these themes are

likely to lend themselves better to some narrative elements

over others. We also found a signicant dierence in com-

prehension for the protagonist (F = 6.02, p = 0.01), which

might be a slight bias in recall for New York generally (the

protagonist in the US sea-level rise story), a populous US

state often dominating media coverage. Signicant dier-

ences in comprehension were not observed for other nar-

rative elements.

Interestingly, participants appeared to have a somewhat

stronger reaction to the US presidential campaign dona-

tions story than the US sea-level rise story, although signif-

icant dierences were not observed for all reaction met-

rics (Table 8). e largest dierence in reaction between

themes regarded participant concern: participants overall

were concerned by the US presidential campaign donation

story (4.9/7), but were neither concerned nor unconcerned

with the US sea-level rise story (4.0/7), a signicant dif-

ference at alpha = 0.01 (F = 14.01, p = 0.00). Participants

also were more upset by the US presidential campaign dona-

tion story (3.5/7) than the US sea-level rise story (2.9/7), a

signicant dierence at alpha = 0.01 (F = 7.24, p = 0.01),

although on average both sets of participants disagreed

with the statement (below 4.0/7) and therefore these rat-

ings should be interpreted as a greater apathy towards the

US sea-level rise story. Participants ultimately agreed more

with the US presidential campaign donation story than the

US sea-level rise story (F = 4.57, p = 0.03), perhaps because

of the increased hedonic reaction to the former story. e

influence of the story theme on participant reaction is

possibly explained by the timing of the study, which was

conducted four months after the polarizing presidential

election of 2016. In contrast, sea-level rise and climate

change broadly work on a longer time scale and within less

well-dened geographic boundaries, a noted challenge for

science communication in getting the public to care about

climate change and its eects (Fish 2020a). Reaction dif-

ferences by theme also may be explained by the scally and

socially liberal lean in the participant sample (Figure 1),

as the US presidential campaign donation story had a subtle

conservative lean (the relationship between political lean-

ing and reaction concern is discussed in more detail under

individual dierences). ere was no statistical dierence

in other reaction metrics.

VISUAL STORYTELLING GENRE (FACTOR 2)

Factor 2 included two conditions by visual storytell-

ing genre: longform infographics and dynamic slideshows.

Unlike themes, overall dierences in retention by genres

were statistically signicant (F = 5.47, p = 0.02) (Table 6).

Participants correctly answered 74.5% of retention ques-

tions for visual stories presented as longform infograph-

ics, but only 68.3% for visual stories containing the same

content but presented as dynamic slideshows. us, it was

the technique used to enforce continuity across narrative

elements (i.e., genres), and not the element content itself

(i.e., themes), that inuenced retention, with participants

Cartographic Perspectives, Number 100, FORTHCOMING Visual Storytelling with Maps – Song et al. | 34

having greater diculty remembering information from

dynamic slideshows. is is an important nding for car-

tography and visual storytelling, as well as science commu-

nication and pedagogy generally, as visual material more

commonly is presented as a discrete slide deck instead of

a continuous graphic or webpage in educational settings

and public presentations. Further, this nding points to

the relevance of emerging design strategies made possible

by new media for map-based visual storytelling, as long-

form infographics enable scrolling at an audience-controlled

pace, whereas dynamic slideshows dose information at a de-

signer-controlled pace. We found signicant dierences

for the easier compare (ordinal) tasks (F = 4.01, p = 0.05)

and more dicult identify (numerical) tasks (F = 8.52, p

= 0.00) tasks, but not for the intermediate-diculty rank

(ordinal) tasks (F = 0.26, p = 0.61), perhaps because of the

aforementioned interaction eect for rank (ordinal) tasks

between theme and tropes.

Regarding comprehension, participants described 49.4%

of the total narrative elements when the story was pre-

sented as longform infographics but only 46.6% when pre-

sented as dynamic slideshows, again indicating better per-

formance with longform infographics. However, this was

not a signicant dierence (F = 0.92, p = 0.34; Table 7),

and overall we observed fewer comprehension dierenc-

es by genres than theme for individual narrative elements.

Interestingly, there was a signicant dierence by genres

in discussion of the problem (F = 6.82, p = 0.01): while

nearly all (99.2%) participants discussed the problem when

viewing longform infographics, only 92.7% of participants

discussed the problem when viewing dynamic slideshows.

Therefore, the dynamic slideshows format and naviga-

tion caused a small set of participants to miss the main

problem altogether in their discussion, a fatal erasure in

comprehension. us, while the retention measures indi-

cated that the dynamic slideshows inhibited recall of spe-

cic numbers for many participants, the comprehension

measures indicated that the dynamic slideshows inhibited

development of a general problem understanding for small

set of participants, together representing the “worst of

both worlds.” One potential explanation for the observed

poorer retention and comprehension for dynamic slideshows

is the manner in which they enforce continuity: clicking

through the frames in dynamic slideshows interrupted the

ow of the story, requiring reloading of new content that

may be exacerbated due to bandwidth lags, altogether

breaking audience concentration. Relatedly, participants

discussed the spatial setting signicantly more frequently

when viewing longform infographics (32.5%) versus dynamic

slideshows (20.3%), possibly because they could more read-

ily access maps—included on only four of nine panels—

through quick scrolling upwards versus reloading a prior

slide. ere was not a signicant dierence in the number

of comprehension mistakes or for other comprehension

codes between genres.

Genres did have a marginal impact on participant reac-

tion (Table 8), although most differences were signifi-

cant only at alpha = 0.10. Participants were more likely to

believe (F = 3.27, p = 0.07) and more likely to have their

views inuenced by (F = 3.14, p = 0.08) stories when pre-

sented as longform infographics versus dynamic slideshows,

an indication of increased inuence and trust evoked by

the emerging “scrollytelling” genre. Notably, participants

reported feeling more upset when viewing dynamic slide-

shows than the longform infographics (F = 6.57, p = 0.01), a

signicant dierence at alpha = 0.05. e negative hedonic

reaction scale did not distinguish between the visual story

content versus design (“I was upset by the visual story”),

and it is possible the unpleasant reaction to dynamic slide-

shows is because of the broken experience of the discrete

slide advance. However, this broken nature also may gen-

erate more suspense and anxiety, a characteristic of dy-

namic slideshows that may be useful when the story topic

and framing require a congruently negative aective re-

sponse (Anderson and Robinson 2022). Finally, dynamic

slideshows were rated as both more exciting (F = 2.76, p =

0.10) and more boring (F = 2.78, p = 0.10) than longform

infographics, a somewhat contradictory nding possibly in-

dicating that dynamic slideshows garner a wider variety of

reactions from the audience.

Notably, Factor 2 was assigned between subjects, unlike

Factors 1 and 3, and therefore the results by genre may

be influenced by potential imbalances in sampling (see

Supplemental Materials). Participants who viewed long-

form infographics reported a greater interest in the inter-

net (6.3/7 versus 5.2/7 for dynamic slideshows), computing

technology (5.7/7 versus 4.5/7), and online news (5.1/7

versus 4.3/7), and a greater familiarity with computing

technology (5.6/7 versus 4.8/7). us, the generally ob-

served benet of longform infographics instead may be ex-

plained by increased participant motivation derived from

greater interest and familiarity with internet and comput-

ing technologies. In contrast, participants who viewed dy-

namic slideshows reported greater training in the internet

(5.2/7 versus 4.4/7 for longform infographics), environment

Cartographic Perspectives, Number 100, FORTHCOMING Visual Storytelling with Maps – Song et al. | 35

and science (4.1/7 versus 2.8/7), online news (4.2/7 ver-

sus 3.0), domestic politics (3.9/7 versus 2.4/7), information

graphics (3.7/7 versus 2.8/7), maps (3.8/7 versus 2.9/7)

and print news (3.4/7 versus 2.6/7). us, maps and vi-

sual storytelling generally were less novel for participants

assigned to dynamic slideshows and therefore this could

have inuenced motivation, prior expectations, and, ac-

cordingly, performance with dynamic slideshows. Sampling

by domain-specic characteristics not captured through

participant demographics is dicult to manage through

Mechanical Turk, and thus future research with purpose-

ful participant recruitment is needed to investigate how

well the observed dierences by genre apply across dier-

ent levels of interest, familiarity, and training.

VISUAL STORY TROPE (FACTOR 3)

Factor 3 included two conditions by visual storytelling

trope, with both conditions employing a visual accenting

technique for focusing attention: leader lines and color high-

lighting. Dierences in retention by tropes were statistical-

ly signicant at alpha = 0.10 (F = 3.51, p = 0.06; Table 6).

Participants correctly answered 73.8% of retention ques-

tions for visual stories using leader lines as the focus atten-

tion strategy, but only 69.1% using color highlighting. While

a weaker inuence on retention than the genres condition

in our study, the dierence by tropes extends the Grin

and Robinson (2015) study in the context of visual accent-

ing between coordinated representations, suggesting that

leader lines are not only a viable alternative to color high-

lighting when color needs to be used elsewhere in design,

but actually are a more salient focusing attention tech-

nique generally. ere were no signicant dierences by

tropes for the easier compare (ordinal) (F = 1.83, p = 0.18)

or rank (ordinal) (F = 1.19, p = 0.28) retention questions,

but differences by tropes for the more difficult identify

(numerical) task were signicant at alpha = 0.10 (F = 3.28,

p = 0.07). us, the benet of leader lines over color high-

lighting—and thus more salient focusing attention tech-

niques—may increase as the diculty in the task grows.

Tropes did not result in a signicant dierence in total

comprehension (F = 0.11, p = 0.74), the number of com-

prehension mistakes (F = 0.29, p = 0.59), or any specic

narrative element (Table 7). We were surprised that none

of the tested factors resulted in signicant dierences in

total comprehension. Although we found the compre-

hension discussion surprisingly rich, with participants on

average describing 48.0% of the narrative elements, the

simplicity of the comprehension question prompt and the

online format of the survey perhaps limited the response

length. Greater depth in comprehension instead may be

elicited through alternative social science methods such as

cognitive walkthroughs, think aloud studies, and post-hoc

interviews and focus groups. However, our method did

present an opportunity to investigate how the visual story

design influenced discussion of the more important or

obvious narrative elements in their shorter qualitative re-

sponses, resulting in the signicant dierences by themes

and tropes for individual narrative elements described

above.

e trope conditions did result in a signicant dierence

in participants’ reported interest in the visual story (F =

4.25, p = 0.04; Table 8). Participants reported being more

interested in visual stories using leader lines (5.2/7) to focus

attention than color highlighting (4.8/7). “Interest’” suggests

an activated aective state, and while we did not observe

a significant difference in positive arousal (which uses

the more suggestive word “excited” to describe reaction),

increased interest in a visual story is a marked benet of

using focusing attention techniques in visual story design:

in this case, leader lines better focused audience attention

on important or unusual information in the story, avoiding

distraction from other design elements or split attention

with other tasks. ere also was a signicant interaction

in reported interest between the genre and tropes factors

(F = 4.81, p = 0.03), indicating that participants were most

interested in visual stories using a combination of longform

infographics and leader lines regardless of the theme. Again,

this is a useful nding for cartographers and data journal-

ists, as the specic theme may not inuence the interest

garnered from a general audience as long as the visual

story follows empirically-derived design recommendations

for a three-act narrative, genres, and tropes. ere were

no addition signicant dierences by tropes in the other

participant reaction scales.

INDIVIDUAL DIFFERENCES

Finally, the exit survey collected a number of measures

of individual dierences, including expertise, motivation,

and prior beliefs. Examination of individual dierences

identied many interaction eects on retention, compre-

hension, and reaction. Our discussion below focuses on

the most prominent patterns.

Cartographic Perspectives, Number 100, FORTHCOMING Visual Storytelling with Maps – Song et al. | 36

Regarding retention, the relationship of individuals to

technology and design was a stronger driver than their re-

lationship to the theme or their prior beliefs (Table 9). In

particular, participants with greater familiarity with the

internet (β= 1.51, p = 0.00) retained signicantly more in-

formation from the visual stories than their counterparts.

is signals an important consideration for data journal-

ism and visual storytelling, as dierential familiarity (i.e.,

expertise) with the online delivery format of the map study

may have impacted results, with participants unfamiliar

with internet technology retaining less information from

the online stories.

ere was an interesting tradeo in retention between fa-

miliarity and interest in print versus online news sourc-

es. Participants with interest in online news sources (β=

1.07, p = 0.00) retained significantly more information

than their counterparts, while participants with interest

in print news sources retained signicantly less than their

counterparts (β= -0.80, p = 0.00). Familiarity with online

news sources (β= -1.20, p = 0.00) and print news sourc-

es (β= 0.57, p = 0.00) reected an inverse pattern. While

future research is needed to track and conrm this eect

by individual dierences, the results possibly suggest that

performance with online visual stories is inuenced by in-

terest in online versus print news sources broadly, but also

suggest a possible tendency to skim visually as familiarity

with online news sources increases.

In contrast, the relationship of individuals to the theme

and their prior beliefs was a stronger relative driver of

comprehension than their relationship to technology and

design (Table 9). As with retention, participants with in-

creased familiarity with the internet (β= 0.71, p = 0.03)

and interest in online news sources (β= 1.05, p = 0.00)

comprehended the visual stories signicantly better than

their counterparts, although this eect was smaller than

with retention. While increased familiarity with on-

line news stories decreased comprehension (β= -0.89, p

= 0.00), similar to retention, there were no observed ef-

fects from familiarity or interest in print news sources on

comprehension.

Instead, increased prior beliefs that sea-level rise is a

topic worth of discussion (β= 0.41, p = 0.03) led to in-

creased comprehension with the US sea-level rise theme.

Interestingly, increased participant training in environ-

ment and science (β= -0.42, p = 0.01) led to reduced com-

prehension about the US sea-level rise theme, suggesting

that participants with some expertise about the theme

actually gleaned fewer specic details from the visual sto-

ries, instead relying on their prior understanding rather

than drawing from specic evidence in the story. is is an

interesting nding, as untrained, non-experts actually may

approach visual stories more objectively, and thus consider

new information and the overall narrative more thorough-

ly, as long as they believe the topic is worthy of discussion.

It is arguable, however, that the subjectivity exhibited by

experts may be an appropriate, hard-earned subjectivity,

allowing them to discredit awed or incomplete informa-

tion and narratives more quickly. Unexpectedly, increased

environmental consciousness and increased concern about

sea-level rise both led to reduced retention and compre-

hension for the US sea-level rise theme, an apparent con-

tradiction to prior beliefs that sea-level rise is a topic worth

discussion.

We did not nd any signicant relationships between prior

beliefs and retention or comprehension of the US presiden-

tial campaign donations condition. is perhaps is due to

the polarizing nature of the 2016 US presidential election,

but broadly indicates that the specic theme matters when

considering individual dierences.

While there are some notable interactions with reten-

tion and comprehension, individual dierences appeared

to have a greater inuence on participant reaction to the

visual stories (Table 10). There were several broad pat-

terns between individual dierences and reaction worth

discussing.

First, familiarity and interest with maps and information

graphics arguably had the largest inuence on the reaction

measures regarding participant relationships to technolo-

gy and design. e more familiar participants were with

maps, the less likely participants were excited by (θ= 0.56,

p = 0.00) or enjoyed (θ= 0.73, p = 0.02) the visual stories,

possibility suggesting that novelty in design plays a role in

the reaction to map-based visual stories. However, greater

interest in maps led to a decrease in boredom (θ= 0.75, p =

0.01) with the visual stories and an increase in enjoyment

(θ= 1.41, p = 0.00), interest (θ= 1.42, p = 0.00), and agree-

ment with the visual stories (θ= 1.27, p = 0.02). Similarly,

greater interest in information graphics led to increased

excitement (θ= 1.35, p = 0.01) and enjoyment (θ= 1.40, p

= 0.01). us, individual dierences in motivation around

data-driven visuals like maps and graphics matter in the

resulting reaction to visual stories containing these maps

and graphics.

Cartographic Perspectives, Number 100, FORTHCOMING Visual Storytelling with Maps – Song et al. | 37

Second, with a few exceptions, familiarity with the story

theme had a greater overall eect on reaction than train-

ing or interest in the topic. As with retention and compre-

hension, greater familiarity with the theme often resulted

in a more negative reaction, as participants reported high-

er boredom (θ=1.33, p = 0.03) and maintenance of their

beliefs about the topic (θ= 1.39, p = 0.01), as well as being

less upset by (θ= 0.76, p = 0.03), interested in (θ= 0.71, p

= 0.01), and concerned about (θ= 0.73, p = 0.02) the visu-

al story when they had greater familiarity with the topic.

While future research is needed to understand this rela-

tionship between expertise with and reaction to visual sto-

ries, these ndings suggest that visual stories may evoke

stronger positive reactions from general audiences less fa-

miliar with the story theme.

ird, as with comprehension, prior beliefs that the story

theme was worth discussing did inuence participant re-

action. Participants more likely to find the topic worth

discussing also were more likely to enjoy (θ= 1.26, p =

0.04) and be interested in (θ= 1.41, p = 0.00) the visual

story as well as less likely to be bored by (θ= 0.77, p = 0.02)

or maintain their beliefs about the visual story (θ= 0.78,

p = 0.03). us, getting audiences to care about the topic

is an important challenge—perhaps the challenge—in vi-

sual storytelling, as the way the audience values the topic

before viewing the map appears to greatly inuence how

they subsequently react to the design and comprehend the

depicted content. Unlike comprehension, this pattern does

hold when reactions are pooled across the US presidential

campaign donations and US sea-level rise themes.

CONCLUSION: IMPROVING THE DESIGN OF MAP-BASED VISUAL STORIES

I  , we reported on an empirical study to un-

derstand and improve the intentional design of map-based

visual storytelling using a case study in data journalism.

Compared to their popularity and wide reach, empirical

research on map-based visual stories remains limited. e

research reported here tackled four emerging design con-

siderations for visual storytelling with maps: story map

themes and their constituent three-act narrative elements,

visual storytelling genres, visual storytelling tropes, and

individual dierences among the audience. Specically,

we asked four research questions to address emerging de-

sign considerations for visual storytelling:

1. What is the inuence of story map themes and their

constituent narrative elements on the audience’s

retention, comprehension, and reaction? Our study

provided initial evidence that a three-act narra-

tive and its constituent narrative elements can be

applied consistently and eectively across visual

story themes, and therefore oers new design

opportunities for cartography and data journalism.

Story map themes did not signicantly inuence

total retention, the “accuracy” or “correctness”

measure used in this study, or total compre-

hension, our alternative performance measure

capturing additional dimensions of engagement

and interest. is nding points to the need for

establishing a research and education agenda on

map-based visual storytelling in both cartography

and data journalism, as the ecacy of some design

decisions are based on not on the story content,

but on the intentional design of the narrative

structure and presentation. Interestingly, partic-

ipants discussed the spatial and temporal setting

signicantly more frequently for the US presiden-

tial campaign donations story and the cause and ef-

fect more frequently for the US sea-level rise story,

potentially indicating that, while themes may not

inuence total comprehension, some individual

narrative elements may be more or less germane

to understanding a given visual story. In contrast,

story themes did inuence audience reaction, with

participants feeling signicantly more concerned

about and upset with the US presidential campaign

donations story, and they ultimately agreed more

with this story, perhaps because of the increased

negative hedonic reaction. While not a direct goal

of our study, this nding reinforces noted chal-

lenges for communication and action on climate

change in the United States.

ere are many opportunities for future research

on narrative structures for map-based storytell-

ing, three-act or otherwise. First, we tested two

visual story themes using the same three-act nar-

rative structure as a control to establish a baseline

for three-act narrative design in cartography and

also mitigate bias from current events or prior be-

liefs given the broader visual story design consid-

erations in the factorial design. More research is

Cartographic Perspectives, Number 100, FORTHCOMING Visual Storytelling with Maps – Song et al. | 38

needed to test the baseline three-act narrative we

described against alternative linear and non-linear

narrative structures, as well as to isolate the in-

uence of specic narrative elements within these

structures to understand how they may be rear-

ranged as narrative anchor points within a visu-

al story. Further, we encourage comparison and

evaluation of a wider range of Vujaković’s (2014)

visual story themes and Phillips’s (2012) visual

story structures, as well as correlation of how well

these designs perform with individual dierenc-

es. Finally, the three-act narrative structure relied

mostly on static visuals, with user interactivity re-

stricted to the genre mechanism for enforcing con-

tinuity in the factorial design (continuous scroll-

ing versus discrete slide advancement). Extensions

with interactivity range from martini-glass and

reverse martini-glass narrative structures plac-

ing interactive maps at the bookends of the visual

story (Segel and Heer 2010) to fully open-ended

and user-controlled interactive digital narratives

(Koenitz et al. 2015).

2. What is the inuence of visual storytelling genres

on the audience’s retention, comprehension, and

reaction? Visual storytelling genres did signicant-

ly inuence total retention. Longform infographics

outperformed dynamic slideshows for both retention

and comprehension, although this dierence was

not signicant for total comprehension. However,

we did observe a signicant dierence in com-

prehension of the problem—the central confron-

tation, obstacle, or setback driving the story that

formed perhaps the most important narrative

element of the story—with nearly all partici-

pants discussing the problem in their open-ended

responses when viewing longform infographics.

Taken together, dynamic slideshows exhibited the

worst of both worlds, causing participants to for-

get specic information (total retention) and, for

some, to miss the overall point of the visual story

(problem comprehension). e poorer performance

with dynamic slideshows likely is attributed to the

manner by which the genre enforces continuity

and doses information, as longform infographics

enabled continuous scrolling at an audience-con-

trolled pace, whereas dynamic slideshows discretely

dosed information in a designer-controlled pace.

is nding also calls into question the optimal

structure for dosing information in both research

and teaching, given that material is commonly

presented as a slideshow for in-person presenta-

tions. While the genre had a weaker inuence on

reaction than the story theme, participants were

more upset with stories presented as dynamic slide-

shows, potentially an aective response attributed

more to the broken nature of the genre structure

than the story content.

Our study suggests that choosing a storytelling

genre as described by Segel and Heer (2010) and

Roth (2021) is a nontrivial design decision, and

thus warrants additional research in cartography,

data journalism, and related elds. We examined

just two visual storytelling genres and future re-

search is needed to fully understand the relative

advantages and limitations of each genre to in-

form their selection, content, and design. Notably,

our ndings contradict those of Wieczorek et al.

(2014) on non-visual text presentation, where pag-

ination (slideshows) outperformed scrolling (long-

form structure). We hypothesize that this could

be attributed either to the inherently two-dimen-

sional nature of maps versus one-dimensional text

or the dierence in control of dosing in our study

versus the increased dosing of the pagination con-

ditions in the Wieczorek et al. study. Regardless,

future research is needed to evaluate different

combinations of text and visuals like maps across

genres, as well as to compare text, visuals, and

other multimedia directly against one another as

alternative story content within any given genre.

As noted above, further research also is needed to

investigate how well the observed dierences by

genre apply across dierent levels of audience in-

terest, familiarity, and training.

3. What is the inuence of visual storytelling tropes

on the audience’s retention, comprehension, and

reaction? As with genres, the focus attention

strategies (one visual storytelling trope) signi-

cantly inuenced total retention, although only

at alpha = 0.10. Retention signicantly improved

when narrative elements were accented by leader

lines instead of color highlighting. While a weak-

er inuence than genres, this nding extends

Grin and Robinson’s (2015) recommendations

for using leader lines for visual accenting across

Cartographic Perspectives, Number 100, FORTHCOMING Visual Storytelling with Maps – Song et al. | 39

multiple representations, as leader lines are not just

an alternative to color highlighting when using color

elsewhere in design, but actually are a more salient

focusing attention technique generally. e benet

of leader lines over color highlighting increased as the

question diculty grew, suggesting more salient

visual accenting techniques are needed as task

complexity increases. As with other factors, there

was not a signicant dierence in total compre-

hension between focus attention strategies, and we

encourage adapting, extending, and triangulating

alternative social science methods in the future to

explore additional dimensions of audience com-

prehension. As with genres, tropes had a weaker

inuence on reaction than the story theme, but

we did observe a signicant dierence in reported

interest by tropes, with leader lines better focus-

ing audience attention on important or unusual

information in the story and avoiding distraction

from other design elements or split attention with

other tasks.

For future research, we encourage evaluation of

the additional visual accenting techniques for fo-

cusing attention in visual storytelling compiled in

Roth (2021), including comparing static solutions

like flow arrows, appended geometric frames,

opacity masks, numbering, call-outs, and labels,

against dynamic solutions like blinking/flicker-

ing, panning/zooming, and focus+context visual-

izations. Further, our attention solution primar-

ily supported elementary tasks similar to Grin

and Robinson (2015), but the leader lines and color

highlighting solutions also should be evaluated

for general map reading tasks, such as provid-

ing overview and summary context for narrative

and exploratory visualization alike (Shneiderman

1996). Finally, we examined continuity (tested

through genres) and attention in our study, hold-

ing all other tropes constant for experimental con-

trol. Future research opportunities exist to explore

and empirically test the roles of mood, dosing, re-

dundancy, metaphor, and voice to support visual

storytelling with maps.

4. What is the inuence of individual audience

dierences on their retention, comprehension,

and reaction? Examining individual dierences

aorded insight into how dimensions of expertise,

motivation, and prior beliefs inuence retention,

comprehension, and reaction. Familiarity with the

internet and familiarity and interest with print

versus online news sources most impacted reten-

tion. While familiarity and interest with online

news sources remained inuential on compre-

hension, prior beliefs that the topic is worthy

of discussion also inuenced comprehension.

Notably, interaction eects regarding prior beliefs

and comprehension were observed for the US

sea-level rise theme only, suggesting that the visual

storytelling theme does matter when considering

individual dierences. Whereas dierences in

genres and trope designs directly inuenced reten-

tion and comprehension—with some variability by

individual dierences—individual dierences, not

the visual story design, appeared to have a greater

overall inuence on participant reaction to the

visual stories. Familiarity with maps and infor-

mation graphics, familiarity with the story theme,

and prior beliefs that the story theme was worth

discussing were among the inuential individual

dierences on participant reaction.

Opportunities for future research on individu-

al dierences in visual storytelling (and cartog-

raphy generally) are numerous. First, our use of

Mechanical Turk, while successful in garnering

wider demographic, geographic, and political di-

versity, limited our ability to assess the inuence

of individual dierences on the tested visual story

design considerations themselves (i.e., by theme,

genres, or tropes). Future research is needed to

replicate our results in a controlled setting with

participants purposefully sampled by individual

differences into comparable groups. Second, we

evaluated just a small range of individual dier-

ences by expertise, motivation, and prior beliefs.

Future research is needed to understand how in-

dividual dierences and intersectional identities

shape understandings from maps and visual sto-

ries (D’Ignazio and Klein 2020), with particu-

lar attention on how our reactions might change

through time around critical events or across

other lived experiences given the increased role

maps play in new stories and data journalism.

Relatedly, and following calls from Kosara and

Mackinlay (2013) and Figueiras (2014), future

research is needed to extend our initial treatment

Cartographic Perspectives, Number 100, FORTHCOMING Visual Storytelling with Maps – Song et al. | 40

of retention, comprehension, and reaction to fully

embrace new measures of engagement, empa-

thy, and emotion resulting from different visu-

al story designs, as the real test of visual story-

telling is its ability to help us make better, more

informed decisions in how we treat ourselves,

each other, and our world. Finally, self-report-

ed, Likert-based measures are a relatively simple

method of capturing information about individual

dierences—necessarily so for the purpose of this

study—and alternative methods that more reliably

screen for interest, familiarity, training, and prior

beliefs are needed to fully explore the role of in-

dividual dierences on the design of map-based

visual stories. This limitation also relates to the

potential issues related to the sampling imbalance

between genre conditions noted above.

ACKNOWLEDGEMENTS

is research was funded by NSF CAREER Grant #1555267 and the Wisconsin Alumni Foundation. We wish to thank

Tanya Buckingham Andersen and Qunying Huang for their feedback on the online map study design and analysis.

REFERENCES

Allen, Mike, ed. 2017. e SAGE Encyclopedia of

Communication Research Methods. Los Angeles: Sage

Publications.

Anderson, Cary L., and Anthony C. Robinson. 2022.

“Aective Congruence in Visualization Design:

Inuences on Reading Categorical Maps.” IEEE

Transactions on Visualization and Computer Graphics

28 (8): 2867–2878. https://doi.org/10.1109/

tvcg.2021.3050118.

Andrienko, Natalia, Gennady Andrienko, and Peter

Gatalsky. 2003. “Exploratory Spatio-temporal

Visualization: An Analytical Review.” Journal of Visual

Languages and Computing 14: 503–541. https://doi.

org/10.1016/s1045-926x(03)00046-6.

Bostock, Mike. 2014. “How To Scroll.” http://bost.ocks.

org/mike/scroll.

Brant, Rollin. 1990. “Assessing Proportionality in

the Proportional Odds Model for Ordinal Logistic

Regression.” Biometrics 46 (4): 1171–1178. https://doi.

org/10.2307/2532457.

Bushell, Sally. 2012. “e Slipperiness of Literary Maps:

Critical Cartography and Literary Cartography.”

Cartographica 47 (3): 149–160. https://doi.org/10.3138/

carto.47.3.1202.

Cairo, Alberto. 2017. Nerd Journalism: How Data

and Digital Technology Transformed News Graphics.

Barcelona: Universitat Oberta de Catalunya.

Caquard, Sébastien. 2011. “Cartographies of Fictional

Worlds: Conclusive Remarks.” e Cartographic Journal

48 (4): 224–225. https://doi.org/10.1179/00087041

1X13203362557264.

———. 2013. “Cartography I: Mapping Narrative

Cartography.” Progress in Human Geography 37 (1):

135–144. https://doi.org/10.1177/0309132511423796.

Caquard, Sébastien, and William Cartwright. 2014.

“Narrative Cartography: From Mapping Stories to the

Narrative of Maps and Mapping.” e Cartographic

Journal 51 (2): 101–106. https://doi.org/10.1179/00087

04114Z.000000000130.

Caquard, Sébastien, and D. R. Fraser Taylor.

2009. “What is Cinematic Cartography?” e

Cartographic Journal 46 (1): 5–8. https://doi.

org/10.1179/000870409X430951.

Cartwright, William. 1999. “Extending

the Map Metaphor using Web Delivered

Multimedia.” International Journal of Geographical

Information Science 13 (4): 335–353. https://doi.

org/10.1080/136588199241238.

Cattoor, Bieke, and Chris Perkins. 2014. “Re-

Cartographies of Landscape: New Narratives in

Architectural Atlases.” e Cartographic Journal 51

(2): 166–178. https://doi.org/10.1179/174327741

3Y.0000000076.

Cartographic Perspectives, Number 100, FORTHCOMING Visual Storytelling with Maps – Song et al. | 41

Caudle, Sharon L. 2004. “Qualitative Data Analysis.”

In Handbook of Practical Program Evaluation, Second

Edition, edited by Joseph S. Whorley, Harry P. Hatry,

and Kathryn E. Newcomer, 417–438. San Francisco:

Wiley

Chapin, Mac, Zachary Lamb, and Bill relkeld. 2005.

“Mapping Indigenous Lands.” Annual Review of

Anthropology 34: 619–638. https://doi.org/10.1146/

annurev.anthro.34.081804.120429.

Cohen, Georey L. 2012. “Identity, Belief, and Bias.” In

Ideology, Psychology, and Law, edited by Jon Hanson,

385–403. Oxford: Oxford University Press.

Cronon, William. 1992. “A Place for Stories: Nature,

History, and Narrative.” e Journal of American History

78 (4): 1347–1376. https://doi.org/10.2307/2079346.

D’Ignazio, Catherine, and Lauren Klein. 2020. Data

Feminism. Boston: MIT Press.

Denil, Mark. 2017. “Storied Maps.” Cartographic

Perspectives 84: 5–22. https://doi.org/10.14714/

CP84.1374.

Eccles, Ryan, omas Kapler, Robert Harper, and

William Wright. 2008. “Stories in Geotime.”

Information Visualization 7 (1): 3–17. https://doi.

org/10.1057/palgrave.ivs.9500173.

Elwood, Sarah. 2006. “Beyond Cooptation or Resistance:

Urban Spatial Politics, Community Organizations, and

GIS-Based Spatial Narratives.” Annals of the Association

of American Geographers 96 (2): 323–341. https://doi.

org/10.1111/j.1467-8306.2006.00480.x.

Engebretsen, Martin, and Helen Kennedy. 2020. Data

Visualization in Society. Amsterdam: Amsterdam

University Press.

Figueiras, Ana. 2014. “How to Tell Stories Using

Visualization.” Paper read at 18th International

Conference on Information Visualisation. https://doi.

org/10.1109/IV.2014.78.

Fish, Carolyn S. 2020a. “Cartographic Content Analysis

of Compelling Climate Change Communication.”

Cartography and Geographic Information Science 47 (6):

492–507. https://doi.org/10.1080/15230406.2020.17

74421.

———. 2020b. “Storytelling for Making Cartographic

Design Decisions for Climate Change Communication

in the United States.” Cartographica 55 (2): 69–84.

https://doi.org/10.3138/cart-2019-0019.

Gershon, Nahum, and Ward Page. 2001. “What

Storytelling Can Do for Information Visualization.”

Communications of the ACM 44 (8): 31–37. https://doi.

org/10.1145/381641.381653.

Gray, Jonathan, Lucy Chambers, and Liliana Bounegru.

2012. e Data Journalism Handbook: How Journalists

Can Use Data to Improve the News: Sebastopol: O’Reilly

Media.

Grin, Amy L., and Julia McQuoid. 2012. “At the

Intersection of Maps and Emotion: e Challenge of

Spatially Representing Experience.” Kartographische

Nachrichten 62 (6): 291–299.

Grin, Amy L., and Anthony C. Robinson. 2015.

“Comparing Color and Leader Line Highlighting

Strategies in Coordinated View Geovisualizations.”

IEEE Transactions on Visualization and Computer

Graphics 21 (3): 339–349. https://doi.org/10.1109/

TVCG.2014.2371858.

Grin, Amy L., Travis White, Carolyn Fish, Beate

Tomio, Haosheng Huang, Claudia Robbi Sluter,

João Vitor Meza Bravo, Sara I. Fabrikant, Susanne

Bleisch, Melissa Yamada, and Péricles Picanço. 2017.

“Designing across Map Use Contexts: A Research

Agenda.” International Journal of Cartography 3 (sup1):

90–114. https://doi.org/10.1080/23729333.2017.1315

988.

Haklay, Muki, Alex Singleton, and Chris Parker.

2008. “Web Mapping 2.0: e Neogeography of

the GeoWeb.” Geography Compass 2 (6): 2011–2039.

https://doi.org/10.1111/j.1749-8198.2008.00167.x.

Haraway, Donna J. 1991. Simians, Cyborgs, and Women:

e Reinvention of Nature: New York: Routledge.

Harley, J. B. 1989. “Deconstructing the Map.”

Cartographica 26 (2): 1–20. https://doi.org/10.3138/

E635-7827-1757-9T53.

Cartographic Perspectives, Number 100, FORTHCOMING Visual Storytelling with Maps – Song et al. | 42

Harrower, Mark, and Benjamin Sheesley. 2005.

“Designing Better Map Interfaces: A Framework for

Panning and Zooming.” Transactions in GIS 9 (2): 77–

89. https://doi.org/10.1111/j.1467-9671.2005.00207.x.

Hauser, David, Gabrielle Paolacci, and Jesse Chandler.

2019. “Common concerns with MTurk as a participant

pool: Evidence and solutions.” In Handbook of

Research Methods in Consumer Psychology, edited

by Frank R. Kardes, Paul M. Herr, and Norbert

Schwarz, 319–337. New York: Routledge. https://doi.

org/10.4324/9781351137713-17.

Hullman, Jessica, Steven Drucker, Nathalie Henry Riche,

Bongshin Lee, Danyel Fisher, and Eytan Adar. 2013.

“A Deeper Understanding of Sequence in Narrative

Visualization.” IEEE Transactions on Visualization and

Computer Graphics 19 (12): 2406–2415. https://doi.

org/10.1109/TVCG.2013.119.

Joliveau, ierry. 2009. “Connecting Real and Imaginary

Places through Geospatial Technologies: Examples

from Set-jetting and Art-oriented Tourism.” e

Cartographic Journal 46 (1): 36–45. https://doi.

org/10.1179/000870409X415570.

Kent, Alexander. 2017. “Trust me, I’m a Cartographer:

Post-truth and the Problem of Acritical Cartography.”

e Cartographic Journal 54 (3): 193–195. https://doi.or

g/10.1080/00087041.2017.1376489.

Koenitz, Hartmut, Gabriele Ferri, Mads Haahr, Diğdem

Sezen, and Tonguç İbrahim Sezen. 2015. Interactive

Digital Narrative: History, eory and Practice. New

York: Routledge.

Kosara, Robert, and Jock Mackinlay. 2013. “Storytelling:

e Next Step for Visualization.” Computer 46 (5):

44–50. https://doi.org/10.1109/MC.2013.36.

Lavery, Matthew Ryan, Parul Acharya, Stephen A.

Sivo, and Lihua Xu. 2019. “Number of Predictors and

Multicollinearity: What are their Eects on Error

and Bias in Regression?” Communications in Statistics-

Simulation and Computation 48 (1): 27–38. https://doi.

org/10.1080/03610918.2017.1371750.

Lidwell, William, Kritina Holden, and Jill Butler. 2010.

Universal Principles of Design, Revised and Updated: 125

Ways to Enhance Usability, Inuence Perception, Increase

Appeal, Make Better Design Decisions, and Teach through

Design. Beverly, MA: Rockport Pub.

Ma, Kwan-Liu, Isaac Liao, Jennifer Frazier, Helwig

Hauser, and Helen-Nicole Kostis. 2012. “Scientic

Storytelling Using Visualization.” IEEE Computer

Graphics and Applications 32 (1): 12–19. https://doi.

org/10.1109/MCG.2012.24.

Macfarlane, Robert. 2007. e Wild Places. New York:

Penguin.

McCullagh, Peter. 1980. “Regression Models for

Ordinal Data.” Journal of the Royal Statistical Society:

Series B (Methodological) 42 (2): 109–127. https://doi.

org/10.1111/j.2517-6161.1980.tb01109.x.

Miller, Christopher C. 2006. “A Beast in the Field:

e Google Maps Mashup as GIS/2.” Cartographica

41 (3): 187–199. https://doi.org/10.3138/

J0L0-5301-2262-N779.

Monmonier, M. 1989. Maps with the News: e

Development of American Journalistic Cartography.

Chicago: University of Chicago Press.

———. 1992. “Authoring Graphic Scripts: Experiences

and Principles.” Cartography and Geographic

Information Systems 19 (4): 247–260. https://doi.

org/10.1559/152304092783721240.

Montello, Dan, and Paul Sutton. 2012. An Introduction

to Scientic Research Methods in Geography and

Environmental Studies. ousand Oaks, CA: SAGE.

Moretti, Franco. 2005. Graphs, Maps, Trees: Abstract

Models for a Literary History. London: Verso.

Muehlenhaus, Ian. 2014. “Looking at the Big Picture:

Adapting Film eory to Examine Map Form,

Meaning, and Aesthetic.” Cartographic Perspectives 77:

46–66. https://doi.org/10.14714/CP77.1239.

Nathans, Laura L., Frederick L. Oswald, and Kim

Nimon. 2012. “Interpreting Multiple Linear

Regression: A Guidebook of Variable Importance.”

Practical Assessment, Research, and Evaluation 17 (1): 9.

https://doi.org/10.7275/5fex-b874.

Cartographic Perspectives, Number 100, FORTHCOMING Visual Storytelling with Maps – Song et al. | 43

Nielsen, Jacob. 2006. “F-shaped Pattern for Reading

Web Content.” Nielsen Norman Group. https://www.

nngroup.com/articles/f-shaped-pattern-reading-web-

content-discovered.

Ormeling, Ferjan. 1995. “New Forms, Concepts,

and Structures for European National Atlases.”

Cartographic Perspectives 20: 12–20. https://doi.

org/10.14714/CP20.890.

Pearce, Margaret Wickens, and Renee Pualani Louis.

2008. “Mapping Indigenous Depth of Place.” American

Indian Culture and Research Journal 32 (3): 107–126.

Pearce, Margaret Wickens. 2008. “Framing the Days:

Place and Narrative in Cartography.” Cartography and

Geographic Information Science 35 (1): 17–32. https://

doi.org/10.1559/152304008783475661.

———. 2009. “Place Codes: Narrative and Dialogical

Strategies for Cartography.” Paper read at 24th

International Cartographic Conference, November 15–21,

Santiago, Chile.

———. 2014. “e Last Piece is You.” e Cartographic

Journal 51 (2): 107–122. https://doi.org/10.1179/17432

77414Y.0000000078.

Phillips, Jonathan. 2012. “Storytelling in Earth

Sciences: e Eight Basic Plots.” Earth-Science

Reviews 115 (3): 153–162. https://doi.org/10.1016/j.

earscirev.2012.09.005.

Riche, Nathalie Henry, Christophe Hurter, Nicholas

Diakopoulos, and Sheelagh Carpendale. 2018. Data-

Driven Storytelling. Boca Raton: CRC Press.

Robinson, Anthony C. 2011. “Highlighting in

Geovisualization.” Cartography and Geographic

Information Science 38 (4): 373–383. https://doi.

org/10.1559/15230406384373.

Rogers, Simon. 2014. “Data Journalism is the New

Punk.” British Journalism Review 25 (2): 31–34. https://

doi.org/10.1177/0956474814538181.

Rose, Gillian. 1997. “Situating Knowledges: Positionality,

Reexivities and Other Tactics.” Progress in

Human Geography 21 (3): 305–320. https://doi.

org/10.1191/030913297673302122.

Roth, Robert E. 2009. “e Impact of User

Expertise on Geographic Risk Assessment under

Uncertain Conditions.” Cartography and Geographic

Information Science 36 (1): 29–43. https://doi.

org/10.1559/152304009787340160.

———. 2013. “An Empirically-derived Taxonomy of

Interaction Primitives for Interactive Cartography and

Geovisualization.” IEEE Transactions on Visualization

& Computer Graphics 19 (12): 2356–2365. https://doi.

org/10.1109/TVCG.2013.130.

———. 2021. Cartographic Design as Visual

Storytelling: Synthesis and Review of Map-based

Narratives, Genres, and Tropes. e Cartographic

Journal 58 (1): 83–114. https://doi.org/10.1080/00087

041.2019.1633103.

Roth, Robert E., and Mark Harrower. 2008. “Addressing

Map Interface Usability: Learning from the Lakeshore

Nature Preserve Interactive Map.” Cartographic

Perspectives 60: 46–66. https://doi.org/10.14714/

CP60.231.

Segel, Edward, and Jerey Heer. 2010. “Narrative

Visualization: Telling Stories with Data.” IEEE

Transactions on Visualization and Computer Graphics

16 (6): 1139–1148. https://doi.org/10.1109/

TVCG.2010.179.

Shneiderman, Ben. 1996. “e Eyes Have It: A Task by

Data Type Taxonomy for Information Visualization.”

Paper read at IEEE Conference on Visual Languages, at

Boulder, CO.

Sieber, Renée E., Pamela J. Robinson, Peter A. Johnson,

and Jon M. Corbett. 2016. “Doing Public Participation

on the Geospatial Web.” Annals of the American

Association of Geographers 106 (5): 1030–1046. https://

doi.org/10.1080/24694452.2016.1191325.

Smith, Jonathan M. 1996. “Geographical Rhetoric:

Modes and Tropes of Appeal.” Annals of the Association

of American Geographers 86 (1): 1–20. https://doi.

org/10.1111/j.1467-8306.1996.tb01743.x.

Cartographic Perspectives, Number 100, FORTHCOMING Visual Storytelling with Maps – Song et al. | 44

Stolper, Charles D., Bongshin Lee, Nathalie Henry

Riche, and John Stasko. 2016. Emerging and Recurring

Data-driven Storytelling Techniques: Analysis of a

Curated Collection of Recent Stories. Redmond, WA:

Microsoft. https://www.microsoft.com/en-us/

research/publication/emerging-and-recurring-data-

driven-storytelling-techniques-analysis-of-a-curated-

collection-of-recent-stories.

Sui, Daniel, and Michael Goodchild. 2011. “e

Convergence of GIS and Social Media: Challenges

for GIScience.” International Journal of Geographical

Information Science 25 (11): 1737–1748. https://doi.org/

10.1080/13658816.2011.604636.

Sweeney, M., M. Maguire, and B. Shackel. 1993.

“Evaluating User-computer Interaction: A Framework.”

International Journal of Man-Machine Studies 38:

689–711. https://doi.org/10.1006/imms.1993.1032.

öny, Matthias, Raimund Schnürer, René Sieber,

Lorenz Hurni, and Renato Pajarola. 2018.

“Storytelling in Interactive 3D Geographic

Visualization Systems.” ISPRS International Journal of

Geo-Information 7 (3): 123. https://doi.org/10.3390/

ijgi7030123.

Vujaković, Peter. 2014. “e State as a ‘Power

Container’: e Role of News Media Cartography

in Contemporary Geopolitical Discourse.” e

Cartographic Journal 51 (1): 11–24. https://doi.org/10.1

179/1743277413Y.0000000043.

Wallace, Timothy R. 2016. “Cartographic Journalism:

Situating Modern News Mapping in a History of

Map-User Interaction.” PhD diss., University of

Wisconsin–Madison.

Weber Reuschel, Anne-Kathrin, Barbara Piatti, and

Lorenz Hurni. 2014. “Data-driven Expansion of

Dense Regions–A Cartographic Approach in Literary

Geography.” e Cartographic Journal 51 (2): 123-140.

https://doi.org/10.1179/1743277414Y.0000000077.

Weber, Wibke, and Hannes Rall. 2012. “Data

Visualization in Online Journalism and its Implications

for the Production Process.” Paper read at 16th

International Conference on Information Visualisation.

https://doi.org/10.1109/IV.2012.65.

Wieczorek, Anna Maria, Zuzanna Kłyszejko, Justyna

Sarzyńska, Agnieszka Szóstek, Karolina Chmiel,

Tomasz Soluch, and Aneta Brzezicka. 2014. “Mode

of Text Presentation and its Inuence on Reading

Eciency: Scrolling versus Pagination.” Studia

Psychologica 56 (4): 309–321.

Young, Mary Lynn, Alfred Hermida, and Johanna Fulda.

2018. “What Makes for Great Data Journalism? A

Content Analysis of Data Journalism Awards Finalists

2012–2015.” Journalism Practice 12 (1): 115–135.

https://doi.org/10.1080/17512786.2016.1270171.